MCP App Store
by KeenEthics
Reporting Ninja icon

Reporting Ninja MCP Server for Claude and ChatGPT

By Reporting Ninja

Overview

Reporting Ninja connects Claude to the marketing data of every client account you have connected in Reporting Ninja: Google Ads, Meta Ads, GA4, Google Search Console, LinkedIn, TikTok, Microsoft Ads, Shopify, HubSpot, Mailchimp and more. Ask in plain words and Claude finds the right account, picks the fields and pulls the numbers from each platform's API: which campaigns spent without converting, how a client is pacing against budget, how this month compares with the last one, or a blended view across channels. With write access switched on by an account owner or admin, Claude can also build the deliverable inside Reporting Ninja: a report from a template or from scratch, in the client's brand, language and currency, delivered by email on a schedule or published in the client portal. What it creates is an ordinary Reporting Ninja report that you can open, edit and send. Read-only until an account owner or admin enables write access. Claude never changes anything in your ad platforms. Deleting a report or sending email to a client always asks for your confirmation first. You need a Reporting Ninja account; sign in as the account owner or an admin.

How to connect

  1. Claude: Open the Reporting Ninja listing in the Claude connectors directory and choose Connect. Claude listing Reporting Ninja exposes 83 tools here.
  2. ChatGPT: Open the Reporting Ninja listing in the ChatGPT app directory and choose Connect. ChatGPT listing Reporting Ninja exposes 83 tools here.

Tools

append_sections

ChatGPT
Append sections to an existing report, with the same composition vocabulary as create_report. The new widgets read from the bindings the report already has (get_report and list_reports publish them); an append cannot add or change data_sources. A template based report, or one whose data sources are per widget, is refused with APPEND_NOT_ALLOWED. Every property of the dsl schema is required, unfulfilled included: send null in the optional ones (chart_type, full_width, content, filter_key, compare, filters, sort_by, limit, date_range, previous_period, show_change, settings) and an empty array in unfulfilled when everything asked for is in the widget. filter_key names a filter SAVED in the account (list_filters) that the whole widget is read through, and it excludes filters, which carries conditions of its own: send one of the two as null. A column of a table_metrics_summary is read through a saved filter of its own by writing that entry of metrics as {"id": "<field id>", "filter_key": "<key>"} instead of the plain field id. compare turns the widget into a COMPARATOR and only a graph_basic, a graph_pie, a graph_trend and a table_metrics_summary take it: by accounts draws one series per connected account and names at least two of them in items, each as {connection_key, account_id, label, filter_key} (list_connections), and such a widget carries no connection of its own -- in a report that binds a connection per widget those items ARE its connections, and sending the connection of the widget as well is refused; by filters draws one per saved filter, each item as {filter_key}, and reads one connection like any other widget; by breakdown splits one page metric of facebook_insights into a series per value -- the metric says what it is broken down by -- and its items are null. The same account or the same filter named twice is refused, and a label cannot carry the sequence "#%#". over_time is the axis a table_metrics_summary comparator is broken down along and show_past_serie the second series of a breakdown; both are null anywhere else, and so are previous_period and show_change, which a comparator does not draw. Which classes and which axes an integration offers depends on the entity level of the data view too. What a widget compares is decided when it is created, and so is whether it is broken down over time at all: update_widget does not change the class of a comparator, does not stop a widget being one and does not switch its axis on or off, and a body with NO compare leaves the comparator exactly as it is. chart_type says which chart a graph_basic (area, line, column, bar) or a graph_trend (area, line, column) is drawn as; null takes the default of the shape -- a column chart for a graph_basic, an area chart for a graph_trend -- and is what every shape that draws no chart sends. full_width true gives the widget a row of its own and nothing else is placed beside it; null, like false, lets it share a row, and a kpi ignores it. previous_period and show_change are the two halves of the comparison a kpi and a table_metrics_summary draw -- the figure of the period before and the percentage of change beside it -- and in those two shapes NULL MEANS YES, which is what they show by default: send false to turn one of them off. In a graph_trend previous_period draws a second series and in a table_dimensions_metrics a column of percentages, and there null asks for nothing; show_change is null in every other shape. visualization text is a block of prose and not a widget of data: it carries its Markdown in content (paragraphs, bold, '- ' and '1. ' lists, at most 2000 characters, no emoji), integration_id is OMITTED for it, every data key is empty or null and it takes a row of its own by default. Take field ids and filter rules from list_fields and describe_integration, and the data_view from the dsl_data_views of the integrations block of get_capabilities (blocks ["integrations"] with integration_id), which is not the data view list of describe_integration. A data_view outside that list is refused. THE LAYOUT OF EACH SECTION IS YOURS …

copy_report

ChatGPT
Duplicate a report under another name, as a report or as a reusable template. The copy carries the sections, the widgets, the data source bindings, the AI summaries with their selections pointed at the new keys, and the AI client briefs. It carries no public link -- two reports cannot share one -- no catalog provenance and no schedules, and it is added to no portal view: a copy is a report nobody is subscribed to yet. To copy a report FOR another client, rebound to that client's connections and optionally scheduled and put in their portal view, use duplicate_report_for_client instead. A report built on a template comes out as a regular report of its own, with copies of the sections it showed, and stops following that template; the answer says so in warnings. as defaults to whatever the source is: a report copied as a report, a template copied as a template. A copy that is a template carries no brand of its own. Turning a report into a template is refused on a report built on a template and on one that assigns a connection per widget; turning a template into a report is refused outright, because that is a materialization. The plan limit is checked before anything is written: with no room left the call is refused with PLAN_LIMIT_REACHED. The answer carries app_url, the link that opens the report in the Reporting Ninja application: give the user that one when they ask for the link, and never build a URL yourself. Opening it requires a signed-in user of the account. It is NOT the public link: share_url is the public one, it is read by anybody holding it, and it exists only once the report has been shared. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_report on the copy. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

copy_section

ChatGPT
Copy one section of a report, with every widget in it, into ANOTHER report of this account. The copy lands at the end of the target report, named after the original with a copy suffix, and each widget of it is pointed at the connections of the report it lands in: in a report that binds one connection per integration they read that report's, and in a report that binds a connection per widget they arrive with none, so follow the call with set_report_connections or set_widget_connection. Use it to carry a section of analysis between two reports instead of describing it again. The source and the target have to be two different reports, and neither of them can be a report built on a template. An AI summary inside the copied section keeps the widgets copied with it and loses any key that named a section or a widget of the report it came from: lost_selections names them. Pass the if_version of the TARGET report to be refused if its sections changed in between. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_section of the section this call created. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

copy_widget

ChatGPT
Copy one widget. The copy takes a row of its own right below the original and everything under it moves one row down, which is what the application does; send to to put it in another section or to open its row somewhere else. The copy carries everything the original has -- its definition and its format, the connection it reads, its notes, and for an AI summary its brief and what it reads -- and it is a widget of its own from then on: changing one does not change the other. Use it to write a variant of a widget that already works instead of describing it again from scratch, then update_widget on the copy. An image and a page break are not copied, the application offers no copy for them either, and a report built on a template is refused: its layout belongs to the template. A summary copied into ANOTHER section loses the widgets it was reading -- they are not there -- and the answer names them in lost_selections. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_widget of the copy this call created. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_brand_template

ChatGPT
Create a brand template of the account: the look reports and client portal views are printed and drawn with -- fonts, colours, margins, the front cover, the table of contents and the header of the portal. A report points at one by naming its key in brand_template (update_report), and several reports share one, so changing the template changes all of them. A NEW TEMPLATE IS BORN COMPLETE: every setting takes the default of the product and the logo of the platform, so a body with a name alone writes a template that prints correctly. Send only what you want to differ. Every field is optional and a null means "leave it as it is", so the object get_brand_template answers can be sent straight back; on a creation an absent field takes the default of the product. The keys a read publishes and a write has no column for -- brand_template_key on a creation, is_system, logo and cover_image -- are accepted and ignored with a READ_ONLY_FIELD_IGNORED warning. Colours are #RRGGBB. THE TWO IMAGES ARE NOT WRITTEN HERE: set_brand_template_logo and set_brand_template_cover_image each take one file. The brand of the platform, under the empty key, is read and never written. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_brand_template. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_calculated_metric

ChatGPT
Create a calculated metric of the account: a column widgets of that integration can draw like any other, and that list_fields publishes afterwards. THREE KINDS, and the body carries the block of one of them: expression is a formula over the metrics of the integration; function aggregates one metric over a day, a week or a month; filtered reads one metric through a saved filter. A formula is a flat list of infix tokens -- metric ids, the operators + - /, parentheses and literal numbers -- and every id has to be one list_fields publishes for that integration; an id that is not is refused by name, which is stricter than the application. integration_id, type and level cannot be changed afterwards. The object get_calculated_metric answers can be sent back as a body: the keys it publishes and a write has no column for -- calculated_metric_key, integration, is_system_defined and the two not_expressible_ -- are accepted and ignored with a READ_ONLY_FIELD_IGNORED warning, the two blocks that are not this metric's kind travel as null, and a null means "leave it as it is". A metric written here is not visible to an application window that is already open: refresh the page before editing it there. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_calculated_metric. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_client_user

ChatGPT
Create a client user of the portal and email them their access. The password is generated by the server, is never published in any answer, and only travels to the address it belongs to. Email addresses are unique across Reporting Ninja: one that already belongs to somebody is refused with USER_EMAIL_EXISTS, so check list_client_users first. The call spends a unit of the account's daily budget of calls that send email. There is no tool that updates a user; delete_client_user removes one. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_client_user. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_email_template

ChatGPT
Create an email template of the account: the email a scheduled delivery is sent as, reused by every schedule that names it with email_template_key. Creating one sends nothing -- only a schedule does -- and the template is not applied to any delivery until one names it. The body and the subject can carry the substitution variables of a delivery, replaced when the email goes out: %%REPORT_TITLE%%, %%CURRENT_PERIOD%%, %%PAST_PERIOD%%, %%EXECUTION_DATE%%, %%CUSTOM_VAR_1%% to %%CUSTOM_VAR_5%% (the substitution_variables of the report), %%REPORT_LINK%% (the link to the report, blank in a subject only when there is none) and %%AI_SUMMARY%% (the text of the report level AI summary, body only). A delivery sent in the link format needs %%REPORT_LINK%% in the body: create_schedule is what checks that, not this tool. sender_email and reply_to_email have to be an address of a user of this account or share the domain of one: this API sends mail on your behalf and the editor is not a rule it can lean on. The body is stored AS IT IS -- it is not sanitized, because a template is a whole HTML document and the platform never paints it; it travels by email. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_email_template. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_filter

ChatGPT
Create a saved filter of the account: a named set of conditions that widgets, sections, portal views, calculated metrics and Data Studio account groups point at by key. USE A SAVED FILTER when the same conditions are meant to be shared and changed in one place; for a filter that belongs to ONE widget, put the conditions in the filters block of its dsl instead, which is what create_widget and update_widget take. A widget that reads a SAVED filter cannot be described by a widget definition today, so update_widget refuses it: what you make here is for the application and for the widgets you do not intend to edit through this API. conditions is the same shape as the filters block of a widget: take the field ids, the operators and whether type_of_field is needed from list_fields and describe_integration, never from memory. For ga4, a saved filter takes standard fields only: the custom dimensions and metrics of a property can filter a widget or a query, not a saved filter, which is not tied to one property. integration_id and level cannot be changed afterwards. The object get_filter answers can be sent back as a body: the keys it publishes and a write has no column for -- filter_key, integration, is_system_defined and the two not_expressible_* -- are accepted and ignored with a READ_ONLY_FIELD_IGNORED warning, and a null means "leave it as it is". A filter written here is not visible to an application window that is already open: refresh the page before editing it there. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_filter. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_image_widget

ChatGPT
Put a picture in a section of a report -- a logo, a screenshot, a diagram -- as one more widget of it. image_base64 is the CONTENT OF A FILE the user provides, encoded by the client that reads it from disk: do not write those bytes yourself and do not generate an image. The file is at most 716800 bytes before encoding and it has to really be a [jpeg, png, gif] -- the first bytes decide, not the name -- and its width and height are read off the picture, never chosen. An image carries no definition, so get_report answers type image with no dsl for it: its picture is replaced with update_image_widget, its title with rename_widget and its place with move_widget. Without a placement it goes to the first row that takes it; an image is not a KPI, so it counts as one of the elements a row holds. A report built on a template is refused. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_widget of the widget this call created. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_language_file

ChatGPT
Create a language file of the account: the literals the widgets, the headers and the field names of a report are printed with, which update_report then puts on a report with language_file. It is written as a WHOLE FILE and not literal by literal -- start from get_language_file of one of the sixteen the platform ships, edit the values, and send the .txt back here base64 encoded. language is required and is what the AI summaries of every report using this file are written in. The answer says ignored_keys: how many keys of the file no field catalogue of the platform knows, dropped exactly as the application drops them. Creating one changes no report until update_report points one at it. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_language_file. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_portal_view

ChatGPT
Create a client portal view: the reports one client sees, in the order of its menu, and the users it is shared with. report_keys come from list_reports and user_keys from list_client_users; the portal_views block of get_capabilities (blocks ["portal_views"]) publishes how many of each a view may carry. With send_notification true the users are emailed the address of the portal, which is why that one call needs a confirmation. confirmation_mode conditional: the call asks for a confirmation when the body carries send_notification true. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_portal_view. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_report

ChatGPT
Create a report (or a template) from a composition: sections copied from the catalog and/or custom sections with the widgets you define, bound to the account's connections through data_sources. Follow the discovery chain first: list_connections for the bindings, list_fields and describe_integration for the field ids and data views of every widget, get_capabilities for the ceilings and the presets. Every property of the dsl schema is required, unfulfilled included: send null in the optional ones (chart_type, full_width, content, filter_key, compare, filters, sort_by, limit, date_range, previous_period, show_change, settings) and an empty array in unfulfilled when everything asked for is in the widget. filter_key names a filter SAVED in the account (list_filters) that the whole widget is read through, and it excludes filters, which carries conditions of its own: send one of the two as null. A column of a table_metrics_summary is read through a saved filter of its own by writing that entry of metrics as {"id": "<field id>", "filter_key": "<key>"} instead of the plain field id. compare turns the widget into a COMPARATOR and only a graph_basic, a graph_pie, a graph_trend and a table_metrics_summary take it: by accounts draws one series per connected account and names at least two of them in items, each as {connection_key, account_id, label, filter_key} (list_connections), and such a widget carries no connection of its own -- in a report that binds a connection per widget those items ARE its connections, and sending the connection of the widget as well is refused; by filters draws one per saved filter, each item as {filter_key}, and reads one connection like any other widget; by breakdown splits one page metric of facebook_insights into a series per value -- the metric says what it is broken down by -- and its items are null. The same account or the same filter named twice is refused, and a label cannot carry the sequence "#%#". over_time is the axis a table_metrics_summary comparator is broken down along and show_past_serie the second series of a breakdown; both are null anywhere else, and so are previous_period and show_change, which a comparator does not draw. Which classes and which axes an integration offers depends on the entity level of the data view too. What a widget compares is decided when it is created, and so is whether it is broken down over time at all: update_widget does not change the class of a comparator, does not stop a widget being one and does not switch its axis on or off, and a body with NO compare leaves the comparator exactly as it is. chart_type says which chart a graph_basic (area, line, column, bar) or a graph_trend (area, line, column) is drawn as; null takes the default of the shape -- a column chart for a graph_basic, an area chart for a graph_trend -- and is what every shape that draws no chart sends. full_width true gives the widget a row of its own and nothing else is placed beside it; null, like false, lets it share a row, and a kpi ignores it. previous_period and show_change are the two halves of the comparison a kpi and a table_metrics_summary draw -- the figure of the period before and the percentage of change beside it -- and in those two shapes NULL MEANS YES, which is what they show by default: send false to turn one of them off. In a graph_trend previous_period draws a second series and in a table_dimensions_metrics a column of percentages, and there null asks for nothing; show_change is null in every other shape. visualization text is a block of prose and not a widget of data: it carries its Markdown in content (paragraphs, bold, '- ' and '1. ' lists, at most 2000 characters, no emoji), integration_id is OMITTED for it, every data key is empty or null and it takes a row of its own by default. Take field ids and filter rules from list_fields and describe_integration, and the data_view from the dsl_data_views of the integrations block of get_capabilities (blocks ["integrations"] with integration_id), which is not the data view list of describe_integration. A data_view outside that …

create_report_from_template

ChatGPT
Create a report (or a template) from a published catalog template. Take the catalog_slug, its current version and the positions of its sections from list_template_catalog with include_sections. Without section_positions the whole template is copied; with them, only those sections and in the order of the template. A parameterized entry needs parameter_values. Bind the template's integrations with data_sources or the widgets will have nothing to read. The answer carries app_url, the link that opens the report in the Reporting Ninja application: give the user that one when they ask for the link, and never build a URL yourself. Opening it requires a signed-in user of the account. It is NOT the public link: share_url is the public one, it is read by anybody holding it, and it exists only once the report has been shared. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_report. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_schedule

ChatGPT
Schedule the delivery of a report by email, through the Reporting Ninja REST API (POST /v1/schedules/create; reference: https://api.reportingninja.com/v1/docs). frequency daily, weekly or monthly with the hour in time; weekly and monthly also need frequency_details (the weekday, or the day of the month). frequency now sends one delivery immediately instead, does not read the time field, and cannot be stopped afterwards. Either name an email_template_key or write the email yourself -- then format, sender, sender_email, reply_to_email, subject and body are all required. The hour is the account's time zone. get_capabilities publishes the frequencies and the recipient ceiling in its schedules block, and the daily budget of calls that send email in guarantees.email_quota (blocks ["schedules", "guarantees"]). confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_schedule. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_section

ChatGPT
Add one empty section at the END of a report. It is born with the settings a new section gets in the application -- its period, the flags of its header, its page orientation and no filters -- and only its name and its page break are asked for; change any of the rest afterwards with update_section. There is no way to ask for another position: use reorder_sections to move it. Widgets go in with create_widget, one at a time, or a whole section at once with append_sections, which writes the section AND its widgets in one call -- prefer that one when you already know what goes in it. A report holds as many sections as it needs. Pass the if_version get_report published to be refused if the sections of the report changed in between; the answer carries the new one. A report built on a template is refused: its sections belong to the template. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_section of the section this call created. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

create_widget

ChatGPT
Add one widget to a section of a report. Take report_key and section_key from get_report, the field ids and filter rules of the dsl from list_fields and describe_integration, and its data_view from the dsl_data_views of the integrations block of get_capabilities (blocks ["integrations"] with integration_id). The widget reads from the report's binding for its integration, so that integration needs a data_sources entry in the report, and a text block needs neither. In a report whose report_level_data_sources is false the widget brings its own connection_key and account_id instead. An AI SUMMARY is written with ai_summary in place of the dsl: the platform writes it for each period, so it has no definition of its own, and the block also carries the settings it is generated with and its brief. Every property of the dsl schema is required, unfulfilled included: send null in the optional ones (chart_type, full_width, content, filter_key, compare, filters, sort_by, limit, date_range, previous_period, show_change, settings) and an empty array in unfulfilled when everything asked for is in the widget. filter_key names a filter SAVED in the account (list_filters) that the whole widget is read through, and it excludes filters, which carries conditions of its own: send one of the two as null. A column of a table_metrics_summary is read through a saved filter of its own by writing that entry of metrics as {"id": "<field id>", "filter_key": "<key>"} instead of the plain field id. compare turns the widget into a COMPARATOR and only a graph_basic, a graph_pie, a graph_trend and a table_metrics_summary take it: by accounts draws one series per connected account and names at least two of them in items, each as {connection_key, account_id, label, filter_key} (list_connections), and such a widget carries no connection of its own -- in a report that binds a connection per widget those items ARE its connections, and sending the connection of the widget as well is refused; by filters draws one per saved filter, each item as {filter_key}, and reads one connection like any other widget; by breakdown splits one page metric of facebook_insights into a series per value -- the metric says what it is broken down by -- and its items are null. The same account or the same filter named twice is refused, and a label cannot carry the sequence "#%#". over_time is the axis a table_metrics_summary comparator is broken down along and show_past_serie the second series of a breakdown; both are null anywhere else, and so are previous_period and show_change, which a comparator does not draw. Which classes and which axes an integration offers depends on the entity level of the data view too. What a widget compares is decided when it is created, and so is whether it is broken down over time at all: update_widget does not change the class of a comparator, does not stop a widget being one and does not switch its axis on or off, and a body with NO compare leaves the comparator exactly as it is. chart_type says which chart a graph_basic (area, line, column, bar) or a graph_trend (area, line, column) is drawn as; null takes the default of the shape -- a column chart for a graph_basic, an area chart for a graph_trend -- and is what every shape that draws no chart sends. full_width true gives the widget a row of its own and nothing else is placed beside it; null, like false, lets it share a row, and a kpi ignores it. previous_period and show_change are the two halves of the comparison a kpi and a table_metrics_summary draw -- the figure of the period before and the percentage of change beside it -- and in those two shapes NULL MEANS YES, which is what they show by default: send false to turn one of them off. In a graph_trend previous_period draws a second series and in a table_dimensions_metrics a column of percentages, and there null asks for nothing; show_change is null in every other shape. visualization text is a block of prose and not a widget of data: it carries its Markdown in content (paragraphs, bold, '- ' and '1. ' lists, at most 2000 characters, no emoji), i…

delete_brand_template

ChatGPT
Delete a brand template of the account. THERE IS NO WAY BACK and the two images go with the row: create_brand_template would write a NEW template with a new key, the logo and the cover image would have to be uploaded again, and every report pointed at the old one would have to be pointed at it. It is REFUSED with IN_USE while any report is printed with it or any client portal view wears it, and details.used_by names them: reports with their key, their internal name and whether they are in the trash -- a report in the trash still points at the template -- and portal_views with their key and name. The answer carries before with every setting of the template, which is what it would take to write it again. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_calculated_metric

ChatGPT
Delete a calculated metric of the account. THERE IS NO WAY BACK: create_calculated_metric would write a NEW metric with a new key, and every widget that drew the old one would have to be changed. It is REFUSED with IN_USE while any widget or Data Studio account group still draws it, and details.used_by names them: reports, each with its key, its internal name, whether it is in the trash and how many widgets and sections of it read the metric, and account_groups, how many of them do. The answer carries before with the whole definition. It does NOT empty the cached results of the account. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_client_user

ChatGPT
Delete a client user of the portal. Only client users are managed here: the key of an owner or an admin of the account answers NOT_FOUND, like a key that does not exist. THERE IS NO WAY BACK: create_client_user would create a new user with a new key and email a new password. REFUSED with IN_USE while the user is in any portal view (details.used_by.portal_views). Nobody is emailed and the user is not told; if signed in at that moment, they keep their session until they sign out or it expires. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_email_template

ChatGPT
Delete an email template of the account. THERE IS NO WAY BACK: create_email_template writes a NEW template with a new key, and every schedule that read the old one would have to be pointed at it. It is REFUSED with IN_USE while any scheduled delivery is sent with it, and details.used_by.schedules names them with their schedule_key, their report_key and their frequency -- point those at another template, or delete them, first. It is refused the same way when it is the template the client portal sends its notifications with (details.reason client_portal), which is changed in the application. The answer carries before with the whole template, body included, which is what it would take to write it again. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_filter

ChatGPT
Delete a saved filter of the account. THERE IS NO WAY BACK: create_filter would write a NEW filter with a new key, and everything that read the old one would have to be pointed at it. It is REFUSED with IN_USE while any widget, section, portal view, calculated metric, alert or Data Studio account group still reads it, and details.used_by names them: reports with their key, their internal name, whether they are in the trash and how many widgets and sections of each read it -- a key there can be a template's, and then there is no internal name to give -- plus calculated_metrics, portal_views, how many alerts and how many account_groups. The answer carries before with the whole definition, which is what it would take to write it again. It does NOT empty the cached results of the account; only update_filter does. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_language_file

ChatGPT
Delete a language file of the account, with all its literals. THERE IS NO WAY BACK: nothing keeps a copy of the .txt, so download it with get_language_file first if it may be wanted again. It is REFUSED with IN_USE while any report of the account is printed with it, and details.used_by.reports names them -- reports in the trash count, because they come back pointing at it. Point those at another file, or at the default of the platform, first. The sixteen files the platform ships cannot be deleted (NOT_ALLOWED). confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_portal_view

ChatGPT
Delete a client portal view. THERE IS NO WAY BACK: the view and its lists of reports and users go for good; the reports and the users themselves are not touched. Clients assigned only to this view can no longer sign in, and the preview names them under users_left_without_view. A client signed in at that moment keeps the view open, with live data, until they sign out or their session expires. Nobody is emailed. create_portal_view writes a NEW view with a new key; before carries what it takes, except a global filter (global_filter says whether there was one). confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_report

ChatGPT
Delete a report with everything inside it. The preview says how many sections and widgets would be lost. By default the report goes to the trash for 7 days: it stops being listed, stops counting against the plan and stops being reachable, and restore_report brings it back with its original type until the 7 days are up. list_trash shows what is in there. permanent true skips the trash and deletes the report for good, and that cannot be undone. Its schedules are deleted either way, and a restore does not bring them back. A report a portal view or a template based report depends on is refused with REPORT_IN_USE, naming what holds it and saying which of those are themselves in the trash. A report already in the trash is only deleted again with permanent true; without it the answer is REPORT_IN_TRASH. A report whose scheduled delivery is being sent right now is refused with SCHEDULE_QUEUED: retry in a few minutes. Retrying with the same idempotency_key returns the original answer even if the report changed state in between: read it with get_report for its current state. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: restore_report. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_schedule

ChatGPT
Stop the scheduled delivery of a report. The row keeps its recipients and its email and stops firing. Take schedule_key from list_schedules. A delivery the scheduler has already been handed cannot be stopped. IT IS ALSO HOW A QUEUED SNAPSHOT IS CANCELLED: send the schedule_key take_snapshot answered (the same one ALREADY_QUEUED repeats) while the snapshot has not been taken, and the request is removed. Nothing lists that key, so it is the one take_snapshot gave you; once the snapshot exists the key answers NOT_FOUND and what you delete is the file, with delete_snapshot. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_schedule with a frequency again. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_section

ChatGPT
Delete one section of a report, with the widgets inside it. The preview says how many widgets would be lost. Take report_key and section_key from get_report, and pass its if_version so the call is refused if the report's sections changed in between. A report built on a template has no sections of its own and is refused; deleting a section of the TEMPLATE removes it, with its widgets, from every report built on it. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. Undo: append_sections re-creates a section from its dsl, but the deleted one is gone -- read it with get_report BEFORE deleting. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_snapshot

ChatGPT
Delete one snapshot of a report, with its file. THERE IS NO WAY BACK: nothing keeps a copy of the file, so the only way to have one again is taking another snapshot, which generates the report as it is today and not as it was. The answer says what the snapshot was. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

delete_widget

ChatGPT
Delete one widget of a section. The preview names the widget that would go. Take the three keys from get_report. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. Undo: create_widget with the dsl get_widget published for it -- read it BEFORE deleting, because afterwards there is nothing left to read. Pass that call's widget_hash as if_widget_hash and the deletion is refused with WIDGET_CHANGED if somebody rewrote the widget in between, which is the one case where the dsl you kept is no longer the widget you are deleting. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

describe_integration

ChatGPT
Describe an integration: its data_views, settings, and supported filters (a static field list with operators, or dynamic dimension/metric operators for GA4). Call list_integrations first to get a valid integration_id.

duplicate_report_for_client

ChatGPT
Copy a report and point it at another client's data, in one call: the copy is made, it is rebound, a schedule is created when you send one, and it is added to a portal view when you name one. It is the whole of what an agency does when it takes on a new client with the same reporting. THE SHAPE OF YOUR BODY FOLLOWS THE SOURCE. A report whose report_level_data_sources is true is rebound with data_sources; one whose report_level_data_sources is false with connections. get_report says which it is. Sending the other one is refused by name, because a rebinding that did nothing would hand the new client a report reading the previous client's accounts. ONLY THE COPY ITSELF CAN REFUSE THE CALL. Once it exists nothing undoes it, so the answer reports the other three steps as they went: bound and failed per binding, schedule {requested, created} and portal_view {requested, added}, each failure with its error_code. Read the answer before telling the user it is done, and fix what failed with set_report_connections, create_schedule or update_portal_view -- never by calling this tool again, which would make a second report. The answer carries app_url, the link that opens the report in the Reporting Ninja application: give the user that one when they ask for the link, and never build a URL yourself. Opening it requires a signed-in user of the account. It is NOT the public link: share_url is the public one, it is read by anybody holding it, and it exists only once the report has been shared. confirmation_mode conditional: the call asks for a confirmation when the body carries schedule or portal_view_key. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_report on the copy. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

duplicate_section

ChatGPT
Copy one section of a report, with every widget in it, INTO THE SAME REPORT. The duplicate lands at the end, named after the original with a copy suffix, and it keeps everything the original had: the connections its widgets read, and what an AI summary inside it was reading, sibling sections included -- they are still in the report. Use it to write a variant of a section that already works, then update_section and update_widget on the copy, or to repeat a block of analysis for another period. To carry a section into ANOTHER report use copy_section instead, which re-binds the widgets to that report's connections. Pass the if_version get_report published to be refused if the sections of the report changed in between. A report built on a template is refused: its sections belong to the template. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_section of the section this call created. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

get_brand_template

ChatGPT
Read one brand template by key: margins, the colour and the font of every part of the report, the palette of the charts, the footer, the cover and the header of the client portal. The empty key is the brand of the platform. The logo and the cover image are reported as present or absent, never carried.

get_calculated_metric

ChatGPT
Read one calculated metric by key: name, the integration and data view it belongs to, its kind and the block of that kind (formula, function or filtered). A widget definition does not take a calculated metric. Which reports and widgets use it is not published here.

get_capabilities

ChatGPT
Return what this account can write and the rules a write has to follow, block by block. Without blocks it returns the default set: contract, widgets (the shapes and how many metrics and dimensions each one takes), composition (the ceilings of a report and its sections, its data sources), date_presets (the periods a report understands), summary, comparators, sources_without_dsl and query_not_available. Every answer also carries account (the plan's room for reports and templates, and whether write access is enabled for this account) and blocks_index, which names every block with one line on what it holds and its size. Any other block comes by its name in blocks: integrations (the dsl_data_views a widget dsl takes its data_view from; with integration_id, that integration alone), filters, calculated_metrics, schedules, portal_views, users, brand_templates, language_files, email_templates, snapshots, reports, editing, versioning, guarantees (confirmation and idempotency lifetimes, undo, the daily email budget, the trash) and endpoints (every write endpoint with its confirmation mode and the tool that undoes it).

get_email_template

ChatGPT
Read one email template of the account by key, WITH its body: the eight fields the application's editor shows -- name, internal notes, format, sender, sender_email, reply_to_email, subject -- and the body itself, which list_email_templates does not publish because it runs to 75.000 characters. Read it before changing it: the object this answers is the body update_email_template takes, and a partial write of a template you have not read is how a body gets replaced by a summary of itself.

get_filter

ChatGPT
Read one saved filter by key: name, the integration and data view it belongs to, whether it is a system one, and its conditions in the shape a widget definition takes -- copy them into a widget rather than naming the filter, which a widget definition cannot do. Which reports and widgets use it is not published here.

get_language_file

ChatGPT
Read one language file by key -- one of the account's own or one of the sixteen the platform ships -- and get a DOWNLOAD LINK for its .txt. The literals themselves never travel in JSON: a language file is six thousand lines. download_url is a signed link that needs no credentials, is good for ONE file and EXPIRES 15 MINUTES after this call (download_url_expires_at says when): hand it to the user, or fetch it, promptly, and call this again for a fresh one. It is the way to get the .txt that create_language_file and update_language_file take: download, edit the values, send it back.

get_query_status

ChatGPT
Check an async query previously submitted with submit_async_query; pass its job_id. status is one of pending, running, success, error. On success, result has the same compact shape as query_data (rows, field_legend, row_count, truncated, next_cursor, ...). On error, error_message explains why. An unknown or expired job_id returns JOB_NOT_FOUND.

get_report

ChatGPT
Return one report in full: its data source bindings, its sections with their section_key, and every widget with its widget_key and, where the widget can be expressed, its dsl (widget-intent-v1) -- the same shape create_widget takes. Widgets that cannot be expressed carry the reason instead. Each section also carries the two keys a section of create_report is written with -- show_previous_period and add_summary -- so a report that is read can be written back without losing them. It also carries the settings of the report -- its brand template, its front cover, its PDF options, its localization, its language file, its substitution variables, what its widgets show, whether its scheduled deliveries store a snapshot, and its AI brief -- under settings, in the very shape update_report takes them, with the names of the brand template and of the language file resolved so naming one needs no second call. That block goes back to update_report as it is: the two resolved names and front_cover.file are ignored there, and a null means the report has no value for that setting. The answer also carries the report's sections version: pass it as if_version on the writes that accept one. A very large report comes back cut on a section boundary with truncated true. A report that delete_report put in the trash is not readable here: the answer is REPORT_IN_TRASH, list_trash shows it and restore_report brings it back.

get_widget

ChatGPT
Return one widget of a report: the same object get_report publishes for it -- what it is, where it reads from and either its dsl (widget-intent-v1) or the named reason it has none -- plus what an edit of it needs. widget_hash is the fingerprint of the definition: send it back to update_widget as if_widget_hash so the call is refused if somebody changed the widget in between. lost_properties says what a rewrite would not be able to keep. backup_keys lists the backups restore_widget can put back, newest first; they expire after 7 days. Call this before update_widget and edit the dsl it answers, instead of writing one from memory. Take the three keys from get_report.

list_brand_templates

ChatGPT
List the brand templates of the account: brand_template_key, name, internal notes, whether it is the brand of the platform, whether it carries a logo and a cover image, its primary colour and its font. Use it to get the brand_template_key update_report takes; the brand of the platform is the empty key. The images themselves are not published.

list_calculated_metrics

ChatGPT
List the calculated metrics of the account: calculated_metric_key, name, the integration and data view they belong to, whether they are system ones, and the block of their kind -- a formula (expression), a metric aggregated over a period (function), or a metric read through a saved filter (filtered). With integration_id it lists only the ones of that integration. They are read and queried with the data tools; a widget definition does not take them.

list_client_users

ChatGPT
List the client users of the account: user_key, email, first and last name, language and creation date. These are the users a portal view can be shared with -- the account's own owner and admins are not listed. Use it to get the user_key values that create_portal_view, update_portal_view and delete_client_user take, and to check whether an email already belongs to a user before creating one.

list_connections

ChatGPT
List the user's connections for an integration, each with its accounts (connection_key + account_id feed list_fields and query_data). For large portfolios the result is bounded: when an integration has more than 200 accounts this returns a compact per-connection summary (account counts + one sample account) instead of every account. Pass connection_key to list one connection's accounts, name_contains to search by account name or id, summary:false to force the full list, or limit/cursor to paginate (default 100, max 500 accounts per call). Read total_accounts / account_count / truncated to know how many exist vs were returned. Call list_integrations first.

list_email_templates

ChatGPT
List the email templates of the account: email_template_key, name, internal notes, whether the body is HTML or plain text, the sender, the reply-to address and the subject. Use it to get the email_template_key create_schedule and update_schedule take. The body is not published.

list_fields

ChatGPT
List the available dimensions and metrics for an integration. Some integrations require data_view (see describe_integration). Narrow large catalogs with type (dimension|metric) and name_contains (substring on field id or name). For per-account custom fields, set include_custom_fields=true and pass connection_key + account_id from list_connections.

list_filters

ChatGPT
List the saved filters of the account: filter_key, name, the integration and data view it belongs to, whether it is a system one, and its conditions in the same shape the filters block of a widget definition takes -- so a filter read here can be pasted into one. With integration_id it lists only the filters of that integration. A filter the API cannot name in public is still listed, with conditions null and the reason why. A widget that reads a saved filter names it in filter_key, of the whole widget or of one column of a table_metrics_summary, and get_widget publishes it there.

list_integrations

ChatGPT
List the advertising/analytics integrations available to your Reporting Ninja account (id, name, category, supports_custom_fields_per_account).

list_language_files

ChatGPT
List the language files a report can be printed with: the sixteen of the platform and the ones the account made itself, each with language_file_key, name, language and whether it is a system one. Use it to get the language_file key update_report takes. The literals themselves are not published.

list_portal_views

ChatGPT
List the client portal views of the account: portal_view_key, name, internal notes, brand template, whether the client may change the date range, the reports of its menu in the order they are shown, and the users it is shared with. Use it to get the portal_view_key that update_portal_view and delete_portal_view take.

list_report_templates

ChatGPT
List the templates the ACCOUNT owns -- created in Reporting Ninja or materialized from the system catalog -- with their report_key, name, number of sections and the connections each one is bound to, the same shape list_reports answers. These are the templates a report of this account can be created from in the app; the system catalog, shared by every account and addressed by catalog_slug, is list_template_catalog.

list_reports

ChatGPT
List the reports of the account -- regular and template based -- with their report_key, name, type, number of sections, the connections each one is bound to and app_url, the link that opens that report in the Reporting Ninja application (give the user that one when they ask for the link of a report; it needs a signed-in user of the account and is not the public share_url). Use it to find the report_key the other report tools take. Templates are two other lists: list_report_templates for the account's own, list_template_catalog for the system catalog. The reports delete_report put in the trash are in none of them: list_trash shows those.

list_schedules

ChatGPT
List the email schedules of the account: schedule_key, the report it delivers, frequency and the day it fires, the hour, the recipient lists, the report format, the email template and where the row came from (app, portal or api). With report_key it lists only the schedules of that report. The body of the email is not published. Use it to get the schedule_key that update_schedule and delete_schedule take.

list_snapshots

ChatGPT
List the snapshots of a report, newest first: the PDF of each delivery that filed one, taken by hand or with a scheduled execution. Each one carries snapshot_key, type (manual or scheduled), status (success or error), the file name, when it was taken and, when it has a file, a download_url. GIVE THAT URL TO THE USER: it is a direct link that needs no credentials, it opens that one file and it expires 15 minutes after this call -- list again for a fresh one. There is no download tool: the file is REST only. Each call MINTS A NEW LINK for every snapshot that has a file, and a link that has been handed out cannot be taken back before it expires: list when the user asks for the file, not to check what is there. A report keeps its 13 newest snapshots.

list_template_catalog

ChatGPT
List the published report templates of the SYSTEM catalog: catalog_slug, version, name, description and the integrations each one reads from. This catalog is the same for every account; the templates the account itself owns are list_report_templates. With include_sections true each entry also lists its sections with their 0-based position -- what create_report addresses in a catalog section and what create_report_from_template takes in section_positions -- and the parameters a parameterized entry asks a value for.

list_trash

ChatGPT
List the reports in the trash of the account, most recently deleted first, with their report_key, name, the type they had before being deleted, when they were deleted and until when they can be restored. Use restore_report to bring one back. These reports are not in list_reports or list_report_templates, and they do not count against the plan.

move_widget

ChatGPT
Move one widget to another row, another column or another section of the same report. Send to with the widget_key of the widget, or layout with several widgets at once: the answer of every move carries previous_layout in the VERY SHAPE layout takes, so passing it straight back as layout puts the report exactly as it was, in ONE call and with nothing to reshape. Every widget the layout does not name keeps the place it has. Rows and columns are numbered from 1 and get_report publishes the ones every widget has. A to without a column puts the widget at the end of the row it lands in, and a row past the last one of the section opens a new row. The rules of the layout are the application's: a row holds up to 4 KPIs or up to 2 other widgets and never both, a widget whose dsl asked for full_width is alone in its row, and a row holding a page break takes nobody else -- a row that does not take the widget is refused with ROW_RULE, naming the widget YOU placed and the first free row of the section. Only the rows this call puts a widget into are judged. A widget that ends up in another section is re-read against that section's date range and filters. A report built on a template is refused: its layout belongs to the template. Send if_position with to to be refused with POSITION_CHANGED if somebody moved the widget in between, and retry with the SAME idempotency_key; it guards the one widget to names and is refused with layout. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: move_widget with the previous_layout this call answered. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

ping

ChatGPT
Health check for the Reporting Ninja MCP server. Returns "pong".

query_blend

ChatGPT
Blend several query results into one table merged on a shared dimension (typically a date). Runs one query per source (2-5 sources) and merges rows by dimension value, labeling each series by its origin; returns rows, a per-series field_legend and the list of sources. Cost: each source is a separate query that counts against your MCP rate limit. Discover ids first with list_integrations -> list_connections -> list_fields. For a single account use query_data instead.

query_data

ChatGPT
Run a report query for one connection/account and return the rows. Follow the discovery chain first: list_integrations -> list_connections (for connection_key + account_id) -> list_fields (for field ids). Provide a date_range (a preset such as {"preset":"lastmonth"}, or {"preset":"custom","start":"YYYY-MM-DD","end":"YYYY-MM-DD"}); optionally compare with compare_to -- each metric is then returned under current/previous/change {absolute,percentage} (an absent change percentage means the previous value was 0; a null value is one that cannot be computed, and its change is null too). Returns up to 1000 rows per call (default 100); when more exist, truncated is true and next_cursor can be passed back as cursor to fetch the next page. Some integrations require data_view, settings or type_of_field on filters (see describe_integration). For large extractions use submit_async_query instead.

rename_widget

ChatGPT
Change the title of one widget, whatever kind of widget it is: a KPI, a chart, a table, a text block, an image, a page break or an AI summary all carry a name and this is how it is changed. Nothing else about the widget is touched, so there is no backup to take and no data is re-read: the undo is another rename with the name get_report published before. Use update_widget when what changes is what the widget SHOWS. The name is between 1 and 80 characters, plain text, and no emoji. In a report built on a template only the text widget the report keeps of its own can be renamed, and the answer says so with affects_only_this_report. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: rename_widget with the previous name. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

reorder_sections

ChatGPT
Put the sections of a report in another order. section_keys carries the SAME keys get_report publishes for the report, each of them once, in the order they are to be read: a list that adds, drops or repeats a key is refused rather than merged, so read the report first and send its keys back rearranged. Pass its if_version to be refused if the sections of the report changed in between. Reordering raises the version of the report and changes nothing about what any widget shows or reads. A report built on a template is refused: its sections, and their order, belong to the template. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: reorder_sections with the previous order. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

restore_report

ChatGPT
Restore a report that delete_report put in the trash, with the type it had before it was deleted. Use list_trash to see what the trash holds and for how long each report stays restorable. A report in the trash does not count against the plan, so the account may be full by now and the restore refused with PLAN_LIMIT_REACHED; a report that is not in the trash is refused with REPORT_NOT_IN_TRASH. The schedules the report had were deleted when it went to the trash and a restore does NOT bring them back: create them again with create_schedule. Retrying with the same idempotency_key returns the original answer even if the report changed state in between: read it with get_report for its current state. The answer carries app_url, the link that opens the report in the Reporting Ninja application: give the user that one when they ask for the link, and never build a URL yourself. Opening it requires a signed-in user of the account. It is NOT the public link: share_url is the public one, it is read by anybody holding it, and it exists only once the report has been shared. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_report. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

restore_widget

ChatGPT
Undo an update_widget: pass the backup_key its answer carried. Every update and every restore keeps the widget as it was first, for 7 days, and get_widget lists what is there under backup_keys when you no longer have the key. The place and the connection of the widget are the CURRENT ones and not the backup's: a widget that was moved stays where it is and one that was rebound keeps reading what it reads. What comes back is what it showed and how it looked. This call takes a backup of its own, so it can be undone the same way. A backup of another widget is NOT_FOUND. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: restore_widget with the backup_key of the answer. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

set_brand_template_cover_image

ChatGPT
Upload the photograph the 'image' front cover of a brand template draws, or remove the one it carries. content_base64 is the CONTENT OF A FILE the user provides, encoded by the client that reads it from disk: do not write those bytes yourself and do not generate an image. At most 716800 bytes before encoding, and JPEG or PNG ONLY -- a GIF is refused with reason format_not_supported, which is what the upload of the application accepts too, and with that same reason whether it is the name or the bytes that give it away. THE FILE IS COMPACTED ON THE WAY IN: what is stored is a copy scaled so its longest side is at most 1600 pixels, aiming at 409600 bytes; a file that cannot be decoded, and one whose copy comes out no smaller, is stored as it was sent. Uploading does NOT change the cover style: set cover_style 'image' with update_brand_template afterwards. REMOVING IT (file null) is refused while the style is 'image', because that cover would have nothing to draw. THE PREVIOUS FILE IS NOT KEPT, so the only way back is uploading it again. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: set_brand_template_cover_image with the previous file. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

set_report_connections

ChatGPT
Point EVERY widget of a report at another connection and another account, one pair per integration. It is the bulk form of set_widget_connection and it is for the reports whose report_level_data_sources is false -- the ones that bind a connection per widget, which get_report says for every report. A report that binds once for the whole report is rebound with the data_sources of update_report and this call is refused on it with APPEND_NOT_ALLOWED and reason report_level_data_sources; a report built on a template is refused with reason template_based. Take connection_key and account_id from list_connections. Each widget is backed up before it is written and its cached results are dropped, so it reads the new account at once. IT IS NOT ATOMIC: each widget is written on its own and the answer lists the ones that were bound and the ones that were not, each of the second kind with its reason -- its integration is not in your body, it reads several connected accounts at once, or its write failed. A widget that reads no data at all (a text block, an image, a page break, an AI summary) is in neither list. Calling again with the same body finishes the job. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: set_report_connections with the previous connections. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

set_report_cover_file

ChatGPT
Upload the PDF the front cover of a report prints, or remove the one it carries. content_base64 is the CONTENT OF A FILE the user provides, encoded by the client that reads it from disk: do not write those bytes yourself and do not generate a PDF. At most 262144 bytes before encoding (524288 on a few accounts) and it has to really be a PDF -- the first bytes decide, not the name, which has to end in .pdf all the same. THE FILE AND THE COVER TYPE ARE WRITTEN TOGETHER and this call resolves the type: uploading moves it to the one that prints the file (none to pdf, report to pdf_and_report), and file null moves it back. Everything else of the cover -- its title, its subtitle, its date line -- is update_report. THE LEGACY COVER TYPES MOVE NOWHERE: a report whose cover is pdf_and_custom takes a REPLACEMENT file and keeps that type, while a file sent to a custom cover, and file null on a pdf_and_custom one, are refused with reason legacy_cover_type -- both would move the cover to the other half of that pair, which the application never does. THE PREVIOUS FILE IS NOT KEPT, so the only way back is uploading it again; the answer names the one it replaced. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: set_report_cover_file with the previous file, or with file null. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

set_widget_connection

ChatGPT
Point one widget at another connection and another account of its own integration. It is for the reports whose report_level_data_sources is false -- the ones that bind a connection PER WIDGET, which get_report says for every report -- and it is refused on a report that binds once for the whole report, where the connection is the report's and is changed there. Take connection_key and account_id from list_connections: they have to be of the same integration the widget already reads, which cannot be changed. Nothing else about the widget changes, the cached results of the widget are dropped so it reads the new account at once, and a backup is taken first -- restore_widget with the backup_key of the answer puts the definition back, and the connection is the one the widget has at that moment. It is refused with WIDGET_NOT_EDITABLE when the widget reads several connected accounts at once, and with INVALID_VALUE when it reads no data at all (a text block, an image, an AI summary). confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: set_widget_connection with the previous connection. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

submit_async_query

ChatGPT
Submit a report query to run asynchronously -- use this for large extractions instead of query_data. Same input as query_data (follow the discovery chain first: list_integrations -> list_connections -> list_fields). Returns a job_id and status=pending; then poll get_query_status with that job_id until status is success or error.

take_snapshot

ChatGPT
Ask for a snapshot of a report: the PDF of it as it is right now, filed with the report. THE FILE IS NOT PRODUCED BY THIS CALL -- the platform generates the report minutes later and the answer is a queued request, not a file. Look for it with list_snapshots afterwards; do not call this again because it has not appeared yet, and do not call it twice in a row: a report that already has one waiting answers ALREADY_QUEUED. IT SPENDS AI CREDITS of the account when the report carries AI summaries -- the preview says how many -- and it sends no email to anybody. A report keeps its 13 newest snapshots; the next one deletes the oldest. confirmation_mode always: every call asks for a confirmation. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: delete_schedule with the schedule_key this call answers, while the snapshot has not been taken; delete_snapshot once the file exists. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_brand_template

ChatGPT
Change the settings of a brand template that already exists. IT IS SHARED: every report printed with it and every client portal view wearing it changes with it, from the next time they are drawn -- list_reports and list_portal_views say which, and delete_brand_template names them when it refuses. The update is PARTIAL: send only the fields you are changing and every other column stays as it is, the logo and the cover image included. The answer carries before with the whole template in the same shape, so sending that block back undoes the call. Every field is optional and a null means "leave it as it is", so the object get_brand_template answers can be sent straight back; on a creation an absent field takes the default of the product. The keys a read publishes and a write has no column for -- brand_template_key on a creation, is_system, logo and cover_image -- are accepted and ignored with a READ_ONLY_FIELD_IGNORED warning. Colours are #RRGGBB. THE TWO IMAGES ARE NOT WRITTEN HERE: set_brand_template_logo and set_brand_template_cover_image each take one file. The brand of the platform, under the empty key, is read and never written. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_brand_template with the before block of the answer. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_calculated_metric

ChatGPT
Change a calculated metric. IT EMPTIES EVERY CACHED RESULT OF THE ACCOUNT, like update_filter and for the same reason: every widget that draws the metric was computed with the old definition. It is expensive on a large account: do not chain it. Deleting a metric does not empty anything. It is a PARTIAL write: what the body does not name is left as it is. The kind, the integration and the data view cannot change, and a metric the platform ships answers NOT_ALLOWED. The answer carries before, which is the body that puts the definition back. The object get_calculated_metric answers can be sent back as it is: the read-only keys are ignored with a warning, the two blocks of the other kinds travel as null and a null means "leave it as it is". A change made here is not visible to an application window that is already open: refresh the page first. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_calculated_metric with the before block of the answer. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_email_template

ChatGPT
Change an email template. EVERY SCHEDULE THAT NAMES IT is sent with the new text from the next delivery onwards -- list_schedules says which ones -- so read it with get_email_template first. It is a PARTIAL write: what the body does not name is left as it is, and the answer carries before, which is the body that puts the template back. It does not send anything and does not empty any cache. The body and the subject can carry the substitution variables of a delivery, replaced when the email goes out: %%REPORT_TITLE%%, %%CURRENT_PERIOD%%, %%PAST_PERIOD%%, %%EXECUTION_DATE%%, %%CUSTOM_VAR_1%% to %%CUSTOM_VAR_5%% (the substitution_variables of the report), %%REPORT_LINK%% (the link to the report, blank in a subject only when there is none) and %%AI_SUMMARY%% (the text of the report level AI summary, body only). A delivery sent in the link format needs %%REPORT_LINK%% in the body: create_schedule is what checks that, not this tool. sender_email and reply_to_email have to be an address of a user of this account or share the domain of one: this API sends mail on your behalf and the editor is not a rule it can lean on. The body is stored AS IT IS -- it is not sanitized, because a template is a whole HTML document and the platform never paints it; it travels by email. The object get_email_template answers can be sent back as it is: every field takes null, and a null leaves the stored value exactly where it is, which is what a read publishes for a column that is empty. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_email_template with the before block of the answer. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_filter

ChatGPT
Change a saved filter. IT EMPTIES EVERY CACHED RESULT OF THE ACCOUNT -- reports and Data Studio alike -- exactly as the application does when somebody saves the filter there, because everything that reads it was computed with the old definition. That is also the way to drop that cache on demand. It is EXPENSIVE on a large account: do not chain it, and do not call it once per condition. Deleting a filter does not empty anything. It is a PARTIAL write: what the body does not name is left as it is, and conditions replaces the whole set. The integration, the data view and the fact that the platform does not own it cannot change: a system filter answers NOT_ALLOWED. The answer carries before, which is the body that puts the definition back. The object get_filter answers can be sent back as it is: the read-only keys are ignored with a warning and a null means "leave it as it is". A change made here is not visible to an application window that is already open: refresh the page first. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_filter with the before block of the answer. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_image_widget

ChatGPT
Replace the picture of an image widget that already exists. image_base64 is the CONTENT OF A FILE the user provides, encoded by the client that reads it from disk: do not write those bytes yourself and do not generate an image. The same rules as create_image_widget -- at most 716800 bytes before encoding, really one of [jpeg, png, gif], and the new width and height read off the picture. Everything else about the widget stays as it is: its name, its row, its column and its section. THE PREVIOUS FILE IS NOT KEPT -- there is no backup of a picture -- so the only way back is uploading the previous one again; the answer says how much the one it replaced weighed. A report built on a template is refused: its pictures belong to the template. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_image_widget with the previous file. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_language_file

ChatGPT
Change a language file of the account. WITH file it REPLACES EVERY LITERAL of the file with the ones of the .txt you send -- there is no partial edit of literals, and the previous .txt is not kept anywhere -- and WITHOUT it not a single literal is touched, which is how the name, the language or the note are changed. Download the current file with get_language_file and edit it rather than writing one from scratch. Every report printed with this file shows the new literals the next time its widgets are drawn: when the file changes the literals the cached results of those reports are emptied, so they are executed again then; nothing is executed by this call, and a change of only the name, the language or the note empties nothing. A window of the application that is already open keeps the previous literals until it is reloaded, and the results it draws meanwhile can be stored again with the old literals and served to the reports printed with the file for up to four hours, so tell the user to reload it. The sixteen files the platform ships cannot be changed (NOT_ALLOWED). It is a PARTIAL write and the answer carries before -- which does NOT carry the literals, so the inverse of a call that sent a file is sending the previous file again. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_language_file with the before block of the answer. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_portal_view

ChatGPT
Change a client portal view. Only the fields the body carries are replaced, and a list it carries replaces the stored one whole, order included -- so report_keys is the new menu and user_keys the new audience, not an addition to them. Take portal_view_key from list_portal_views. confirmation_mode conditional: the call asks for a confirmation when the body carries any of report_keys, user_keys or send_notification. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_portal_view with the previous values, which the preview of a confirmed call shows before and after. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_report

ChatGPT
Change the settings of a report that already exists: its title, its subtitle, its internal notes, its public link, the connections it reads through, the period of every one of its sections, and the whole settings window -- brand template, front cover, PDF options, localization, language file, substitution variables, what the widgets show, whether scheduled deliveries store a snapshot, and the AI brief of the report. It is a PARTIAL update -- only the fields you send are written, and inside a block only the keys you send -- so everything you leave out is left exactly as it is. get_report publishes all of them under settings, in this same shape. date_range is the one that closes a month: the period belongs to each SECTION, and this field moves ALL of them to the same range in one call, which is what the button of the application does. It writes the range and nothing else, so each section header goes on showing whatever it showed, and the cached results of the report are dropped so the next read is of the new period. The answer lists the sections it moved. data_sources rebinds a report whose report_level_data_sources is true (get_report says so for every report); a report that binds a connection per widget is rebound with set_report_connections instead, and data_sources is refused on it by name. share is the public /share link: the id is minted the first time the report is shared and again whenever regenerate is true, which kills the link that was given out. A template has no public link. localization either follows the account or belongs to the report, which is what localization.use says: the six settings under it exist only in the second case, sending one of them with use account is refused, and moving a report to its own localization asks for the six together. front_cover.type commands the file: a type that prints an uploaded PDF is refused on a report that carries none, and a type that prints none removes the file the report carried (the answer says so as a COVER_FILE_REMOVED warning, a dry run included). Changing the language file, the localization mode or the date format drops the cached results of the report, so the next read is of the new setting; nothing else does and nothing is executed by the call. The settings block get_report answers is this body, whole: send it straight back and it writes what the report already has. A key carrying null means the column has no value and is left alone; an empty string is a value. brand_template_name, language_file_name and front_cover.file are published by the read and written by nothing here: they are accepted, ignored, and named in a READ_ONLY_FIELD_IGNORED warning. The uploaded PDF of the cover is the one thing of that window this call does not write -- it is a file, and set_report_cover_file uploads it and removes it, writing the cover type with it. What no call of this API will ever change is the LEVEL at which a report binds its data sources. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_report with the values the preview showed under before. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_schedule

ChatGPT
Change an existing schedule, through the Reporting Ninja REST API (POST /v1/schedules/update; reference: https://api.reportingninja.com/v1/docs). Only the fields the body carries are replaced, and a list it carries replaces the stored one whole, order included. Take schedule_key from list_schedules. Rows that carry settings this version does not publish -- conditional execution, the system email template -- are refused with NOT_EDITABLE_BY_API; those can still be stopped with delete_schedule. confirmation_mode conditional: the call asks for a confirmation when the body carries any of recipients_to, recipients_cc, recipients_bcc, frequency, frequency_details, time, report_format, email_template_key, format, sender, sender_email, reply_to_email, subject or body. Call once; if the answer starts with CONFIRMATION_REQUIRED, show the preview to the user, get an explicit yes, and call again with the SAME arguments and the SAME idempotency_key plus confirmation_token. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_schedule with the previous values, which the preview of a confirmed call shows before and after. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_section

ChatGPT
Change the settings of one section of a report: its title, what its header shows, how its pages are laid out in the PDF and the period its widgets read. Send only the fields you are changing -- everything else the section holds is kept -- and the answer carries a before with the previous value of each one, which is what undoes the call. date_range moves THIS section only, and every widget of it that does not carry a period of its own; update_report with a date_range moves every section of the report at once, which is how a month is closed. A section that does not show its date range cannot show the period it is compared against, because the one is drawn under the other: asking for both writes show_previous_period false and says so in warnings. The filters and the Analytics segment a section applies are not written by this version. A report built on a template is refused: its sections belong to the template, where changing one changes every report built on it. confirmation_mode never: there is no confirmation step; the write happens on the first call. idempotency_key is required: a fresh UUID per operation, reused on a retry and in the confirmation step. dry_run true validates everything and writes nothing. Undo: update_section with the values of before. Writing needs an account owner or admin to have enabled write access in the AI connector (MCP) section of Reporting Ninja.

update_widget

ChatGPT
Fix a widget that already exists instead of deleting it and creating another one. Call get_widget first: send back its dsl with your change applied, and its widget_hash as if_widget_hash. The WHOLE definition travels, not the part that changes -- the widget is rebuilt from it. An AI SUMMARY has no definition: it is rewritten with ai_summary, which changes what it reads, the settings it is generated with and its brief. The settings and the brief are a partial write -- what the block does not name is left as it is -- while the SELECTION is resolved on every call: send widget_keys or section_keys back as get_widget published them, or the platform picks what the summary reads again -- and in scope widget widget_keys is required, so it travels even when what changes is a setting or the brief. The rewrite KEEPS what the definition does not say: the connection the widget reads through, the row and the column it is in, its format while its shape does not change, the literals of a custom period header, the title of every column whose metric is still there, and the internal notes unless internal_notes is sent. What it cannot keep comes back in lost_properties. The integration a widget reads from cannot be changed, so there is no integration_id here, and full_width is not read: an update never moves the widget. Every property of the dsl schema is required, unfulfilled included: send null in the optional ones (chart_type, full_width, content, filter_key, compare, filters, sort_by, limit, date_range, previous_period, show_change, settings) and an empty array in unfulfilled when everything asked for is in the widget. filter_key names a filter SAVED in the account (list_filters) that the whole widget is read through, and it excludes filters, which carries conditions of its own: send one of the two as null. A column of a table_metrics_summary is read through a saved filter of its own by writing that entry of metrics as {"id": "<field id>", "filter_key": "<key>"} instead of the plain field id. compare turns the widget into a COMPARATOR and only a graph_basic, a graph_pie, a graph_trend and a table_metrics_summary take it: by accounts draws one series per connected account and names at least two of them in items, each as {connection_key, account_id, label, filter_key} (list_connections), and such a widget carries no connection of its own -- in a report that binds a connection per widget those items ARE its connections, and sending the connection of the widget as well is refused; by filters draws one per saved filter, each item as {filter_key}, and reads one connection like any other widget; by breakdown splits one page metric of facebook_insights into a series per value -- the metric says what it is broken down by -- and its items are null. The same account or the same filter named twice is refused, and a label cannot carry the sequence "#%#". over_time is the axis a table_metrics_summary comparator is broken down along and show_past_serie the second series of a breakdown; both are null anywhere else, and so are previous_period and show_change, which a comparator does not draw. Which classes and which axes an integration offers depends on the entity level of the data view too. What a widget compares is decided when it is created, and so is whether it is broken down over time at all: update_widget does not change the class of a comparator, does not stop a widget being one and does not switch its axis on or off, and a body with NO compare leaves the comparator exactly as it is. chart_type says which chart a graph_basic (area, line, column, bar) or a graph_trend (area, line, column) is drawn as; null takes the default of the shape -- a column chart for a graph_basic, an area chart for a graph_trend -- and is what every shape that draws no chart sends. full_width true gives the widget a row of its own and nothing else is placed beside it; null, like false, lets it share a row, and a kpi ignores it. previous_period and show_change are the two halves of the comparison a kpi and a table_metrics_summary draw -- the figure of the period before and…

append_sections

Claude

copy_report

Claude

copy_section

Claude

copy_widget

Claude

create_brand_template

Claude

create_calculated_metric

Claude

create_client_user

Claude

create_email_template

Claude

create_filter

Claude

create_image_widget

Claude

create_language_file

Claude

create_portal_view

Claude

create_report

Claude

create_report_from_template

Claude

create_schedule

Claude

create_section

Claude

create_widget

Claude

delete_brand_template

Claude

delete_calculated_metric

Claude

delete_client_user

Claude

delete_email_template

Claude

delete_filter

Claude

delete_language_file

Claude

delete_portal_view

Claude

delete_report

Claude

delete_schedule

Claude

delete_section

Claude

delete_snapshot

Claude

delete_widget

Claude

describe_integration

Claude

duplicate_report_for_client

Claude

duplicate_section

Claude

get_brand_template

Claude

get_calculated_metric

Claude

get_capabilities

Claude

get_email_template

Claude

get_filter

Claude

get_language_file

Claude

get_query_status

Claude

get_report

Claude

get_widget

Claude

list_brand_templates

Claude

list_calculated_metrics

Claude

list_client_users

Claude

list_connections

Claude

list_email_templates

Claude

list_fields

Claude

list_filters

Claude

list_integrations

Claude

list_language_files

Claude

list_portal_views

Claude

list_report_templates

Claude

list_reports

Claude

list_schedules

Claude

list_snapshots

Claude

list_template_catalog

Claude

list_trash

Claude

move_widget

Claude

ping

Claude

query_blend

Claude

query_data

Claude

rename_widget

Claude

reorder_sections

Claude

restore_report

Claude

restore_widget

Claude

set_brand_template_cover_image

Claude

set_report_connections

Claude

set_report_cover_file

Claude

set_widget_connection

Claude

submit_async_query

Claude

take_snapshot

Claude

update_brand_template

Claude

update_calculated_metric

Claude

update_email_template

Claude

update_filter

Claude

update_image_widget

Claude

update_language_file

Claude

update_portal_view

Claude

update_report

Claude

update_schedule

Claude

update_section

Claude

update_widget

Claude

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/reporting-ninja"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/reporting-ninja)