MCP App Store
by KeenEthics
Reach Reporting icon

Reach Reporting MCP Server for ChatGPT

By Reach Reporting

Overview

Reach Reporting turns your accounting data into finished, presentation-ready financial reports through conversation. Describe the report you want in plain language - the metrics, the audience, the period - and Reach reads your company's real financials, designs the pages, builds the report, and gives you a preview to open. Keep refining by talking: make revenue a bar chart, add a cash flow page, use our brand colors. Nothing is published until you explicitly approve it. You can also just ask questions of your books: profit and loss, balance sheet, cash flow, ratios and budget variance, scoped to any company you already have permission to view in Reach. Periods resolve against each company's closed accounting months rather than today's calendar date, so the figures reconcile with the books. Reach reports on your accounting data and cannot move money, initiate payments or transfers, or alter ledger transactions. One connection covers your whole firm - every company you can view in Reach is available, and clients added later appear automatically. Requires a Reach Reporting account with at least one AI-enabled company.

How to connect

  1. ChatGPT: Open the Reach Reporting listing in the ChatGPT app directory and choose Connect. ChatGPT listing Reach Reporting exposes 79 tools here.

Tools

add_image_to_library

ChatGPT
Add an image to the signed-in user’s Reach Image Library from a public URL, so it can be placed on a report page. Reach downloads the image itself — pass the direct URL of the image file and nothing else; do not try to send the bytes. Returns { id, name, url, width, height }; use the url as the href of an image element and size the element from width and height. PNG and JPEG only, up to 10MB, over https. The URL must point straight at the image file, not at a page that displays it. For an image that does not exist yet, or one that exists only on the user’s own machine, this tool cannot help — ask the user to upload it in Reach. For a flat shape or a simple mark (a swatch, divider, rule, badge or block) do not use an image at all: draw it with the shape elements, which stay sharp at any size and follow the report’s colours. Requires company_id (from list_companies); the image is stored against the signed-in user.

append_statement_pages

ChatGPT
Append the locked Profit & Loss, Balance Sheet, and Cash Flow statement pages to the END of an EXISTING CANVAS report (Cash Flow only when the accounting integration supports it). The pages are generated with the report's AI styling — do NOT author statements yourself as tableKpi. Enqueues a job and returns a job_id — poll get_ai_job_status until it completes. Requires company_id, report_id, and range (LAST_MONTH, LAST_FISCAL_QUARTER, or FISCAL_YEAR_TO_LAST_MONTH); pass budget_type/budget_id to compare against a budget. By default every statement type the integration supports is appended; pass statement_types to append only a subset (e.g. just the Balance Sheet). The same statement type can appear more than once at different ranges — call again with a different range to add, e.g., a YTD P&L (FISCAL_YEAR_TO_LAST_MONTH) alongside a monthly one (LAST_MONTH). Confirmed reports must be locked first with lock_report. If the report already has statement pages of a requested type AT THIS RANGE the call errors unless replace=true, which deletes and regenerates only those; pages of a type not requested, or the same type at a different range, are left untouched. Authored pages are unaffected and the authored-page cap does not count statement pages.

confirm_report

ChatGPT
Confirm (publish) a temporary design-session report: set temporary=false so it appears in list_reports and later modify_report edits use the tableKpi position-only lock. Call ONLY after the user has reviewed via open_report_print_preview and given an explicit yes to publish — never automatically at the end of design_report / create / update, and never before presenting the print preview. This tool renders a UI widget (a clickable button) in the calling turn — if you are an orchestrator delegating report work to a subagent, do NOT let the subagent call this tool; have it stop short and return report_id/company_id to you, then call this yourself in the top-level conversation so the button actually renders for the user. On success returns a report editor url (open_in_browser) — present/open that link (not print preview). Args: company_id, report_id. This is NOT create_report_from_svg/update_report_from_svg finalize= (that flag only ends a multi-page SVG draft job). Idempotent if already confirmed.

create_dashboard_from_ir

ChatGPT
Build a NEW Reach dashboard from grid-native dashboard IR, in ONE call carrying the whole surface — a dashboard is a single grid, so there is no page chain and no draft to accumulate. Takes company_id, dashboard_name and tiles (max 60): table tiles bound to live data, and divider tiles carrying section headings. Every tile carries {x, y, w, h} in whole grid cells against get_dashboard_capabilities — call that first; pixel geometry is rejected. Returns job_id; poll get_ai_job_status. The dashboard publishes as soon as it builds, so there is no print preview and no confirm step. Resending the same dashboard returns the same job_id instead of building a second one.

create_datasheet

ChatGPT
Create a company datasheet (not Google/Excel, not a budget datasheet). Pass company_id, name, optional start/end as YYYY-MM (defaults to the current UTC calendar year), and optional mixed rows in display order. CUSTOM: { title, values?: { "2026-01": 100 } } or { title, formula: "={Revenue}-{COGS}" } using {Title} / {Title, -1} / {Title, YYYY-MM} (compiled to Excel in customValues). ACCOUNT: { item_type, identifier, title? } from tableKpi binds (account, accountTotal, cfAccount, total, datasheetRow, googleSheetRow, excelSheetRow). Optional number_format (percent|numeric|currency), decimals (0-7), and per-month cell_formats / cell_decimals. Percent values are fractions (0.5 displays as 50%). Does not add the UI 10 blank placeholders. Returns { id, name, startDate, endDate, months, rows: [{ id, title, type, values, item?, formula?, number_format?, cell_formats?, decimals?, cell_decimals? }] }. Edit with update_datasheet; read with list_datasheets / list_datasheet_rows / get_datasheet. Caps: 200 CUSTOM + 200 ACCOUNT, 60 months.

create_image_upload_url

ChatGPT
Get a URL you can PUT image bytes to yourself, for when YOU hold the file — one you generated, or one the user attached to this conversation that you can read in your own environment. Send the bytes with an HTTP PUT to the returned upload_url, then call save_uploaded_image with the upload_id to put it in the user’s Image Library and get back a url you can place as an image element href. Only .png, .jpg and .jpeg. If you cannot make outbound HTTP requests, or the file is on the user’s machine and not yours, use request_image_upload instead and hand them the link. For an image already on the web, use add_image_to_library with its URL. Requires company_id (from list_companies).

create_metric

ChatGPT
Create a new KPI (tableKpi) on an existing CANVAS report page. Pass company_id, report_id, report_page_id, title, view (table|column|line|bar|pie|scatter|waterfall), and IR rows/columns from get_metric_capabilities (views) and get_report_capabilities / get_report_context (ACCOUNT binds + date/forecast columns with range/rangeOffset; custom columns have no dates; title on custom by default, other types only if the user asks). Optional rows[].selected and columns[].selected hide or show; omit means shown. CUSTOM rows: formula ={Title}-{COGS} or values into customValues. ACCOUNT values (=Excel) only under custom columns. Excel functions: get_formula_capabilities. Optional series_colors and geometry {x,y,width,height} (default: full content width, top 300, height 300). Optional formatting_rules (cells only: rows+columns+op+value → color/icon; icon_color defaults to #333333). On confirmed reports call lock_report first — if this user already has the report open in Reach, lock_report takes over (read-only + live updates); then create_metric; call unlock_report when done. Refuses locked statement pages and client portal. Caps: 30 rows, 24 columns — a focused KPI, not a full statement. Returns table_id for update_metric / get_metric. Inspect cellStyles and formatting_rules with get_metric. Open canvas editors live-add via newMetricCreated.

create_report_from_ir

ChatGPT
Build a CANVAS report from Reach IR JSON — the structured payload SVG compiles to (schema: reach-svg-ir), ~45% smaller than tagged SVG with no escaping. ONE PAGE PER CALL (max 12). First call: company_id, report_name, page, finalize=false → returns draft_id. Subsequent pages: draft_id + page. Last page: finalize=true (+ page) with range → job_id; or a follow-up with draft_id + finalize=true + range and no page to finalize pages already stored. Pass budget_type/budget_id when the report compares to budget. Poll get_ai_job_status, then confirm_report. Do NOT pass multiple pages in one call. A page identical to the last stored one is treated as a retry: it is not added again, page_count is unchanged, and the response carries deduplicated:true — to add a genuinely repeated page, change its content first.

delete_report_page

ChatGPT
Remove one page from a report. This is a SOFT delete: the page is hidden and no longer rendered or listed, its data tables are left intact, and it can be recovered later. Call ONLY after the user has explicitly said to remove this specific page — never speculatively, and never automatically as part of an edit loop. Refuses to remove the last remaining page (a report needs at least one). Works on temporary and confirmed CANVAS reports. Requires company_id, report_id, report_page_id (from get_report_contents). To remove a page you just added by mistake, this is the tool; to change page positions use reorder_report_pages.

design_dashboard

ChatGPT
Start an interactive session to design and build a NEW Reach dashboard with the user. Use when the user wants a dashboard — NOT a report (use design_report) and NOT editing an existing dashboard. Requires company_id (from list_companies or the user). Mutates nothing — it returns the questions to ask and the step order to follow, going straight to the brief; the grid contract comes from get_dashboard_capabilities.

design_report

ChatGPT
Start an interactive session to design, build, or lay out a NEW Reach report with the user. Use when the user wants to design/build/create a report — NOT the one-shot create_ai_report job, and NOT editing an existing report (use modify_report for that). Requires company_id (from list_companies or the user). Mutates nothing — it returns the questions to ask and the step order to follow, opening with a how-to-start pick — from scratch (the default), from a curated sample, or from a PDF the user attaches. list_report_samples is called only once the sample route is chosen, never before that pick; a chosen sample is then fetched with get_report_sample and adapted to the company.

duplicate_report_page

ChatGPT
Duplicate one existing authored (CUSTOM) page — copies its layout and its data tables — and place the copy directly after the source page. Locked statement pages cannot be duplicated. Use this for "add another page like that one" or to fork a page before editing a variant. The copy is a real new page with its own id; read it back from get_report_contents, then edit it with update_report_from_svg. To add a blank/new-content page from an SVG instead, use update_report_from_svg. Requires company_id, report_id, report_page_id (from get_report_contents).

get_ai_date_columns

ChatGPT
Resolve the AI Generator date-column sets for a closed AiRange anchored on the company book month (not calendar today). input_columns are windows to analyze when deciding what to present, with optional role tags (current|prior_period|prior_year|forecast). comparison labels those windows and supplies default_basis (prior_period or prior_year) plus live_text_range2 (PREVIOUS_PERIOD or PREVIOUS_YEAR) for actuals analysis. For a budget pair BOTH are null, as prior_period and prior_year are — a budget variance sets liveText range2 = range, not a prior period — and comparison.forecast holds the plan window (same dates as current). output_columns are for trend metrics (KPI; SVG tableKpi) via data-reach-tableKpi-columns; change/variance metrics use comparison prior+current date columns plus custom columns with formula CHANGE and PERCENT_CHANGE (actuals) or date+forecast plus a custom CHANGE column (budget). A custom column without formula does not self-compute. Each date/forecast column and comparison window includes range (named key) and rangeOffset — copy both onto tableKpi columns so periods stay live as the book month advances; do not author startDate/endDate alone. For fiscal-quarter ranges, rangeOffset must be a multiple of 3 (0, -3, -12); do not reuse a LAST_MONTH offset such as -1 on LAST_FISCAL_QUARTER. live_text_range is the user-selected AiRange key for liveText. book_month is the closed book month (the month Last Month names) and the anchor for explaining periods. Call after design_report collects data type + range (+ budget when comparing). Pass budget_type + budget_id together for Actuals vs Budget/Forecast; omit both for actuals.

get_ai_job_status

ChatGPT
Poll status for a create_report_from_svg or update_report_from_svg job. Returns waiting/active/completed/failed. When completed successfully, includes report_id for the CANVAS report.

get_balance_sheet

ChatGPT
Balance sheet; same filters as get_profit_loss. Response shape matches get_profit_loss (months + rows[].values;). Requires company_id (from list_companies or the user).

get_book_month

ChatGPT
The company's closed book month (YYYY-MM-DD UTC), used as the end anchor for relative ranges like 'last N months' (book month plus N-1 prior months). Call this when the user does not give explicit months. Requires company_id (from list_companies or the user).

get_cash_flow

ChatGPT
Cash flow; same filters as get_profit_loss. Same response shape: months + rows[].values. Requires company_id (from list_companies or the user).

get_chart_of_accounts

ChatGPT
Chart of accounts. Response: accounts (flat, max 500) with id, name, parentId, type (enum key or null), subtype (enum key or null), cashFlowSection (enum key or null). Statement rows use node.value like account_{accountId}_{companyId} for GL lines — but on account-mapped companies leaf lines are map_{uuid}_{companyId} and id is the uuid; rollups are accountTotal_{accountId}_{companyId} or mapTotal_{uuid}_{companyId}; section totals are total_income_{companyId}. The bindable identifier is always the bare middle segment (numeric id or uuid); author itemType account/accountTotal and compile flips it to map/mapTotal for mapped companies. Pass leaf account ids via account_ids, parent/rollup ids via accountTotal_ids, or section tokens (e.g. income) via total_ids. Requires company_id (from list_companies or the user).

get_dashboard_capabilities

ChatGPT
The dashboard grid contract for laying out a NEW dashboard: column count, row height, minimum tile size and placement rules in grid cells, plus the closed list of tile types. Call once at the start of a design_dashboard session. Not reports — those use get_report_capabilities.

get_dashboard_contents

ChatGPT
Read an existing dashboard as editable IR: every tile with its data bindings and its {x, y, w, h} grid cells. Returns tiles in the same shape create_dashboard_from_ir takes, so you can change one and submit the result. Tiles this grammar cannot express — live texts, query tables, AI widgets, locked statements — come back under unsupported: pass them back exactly as given so the edit leaves them alone. Requires company_id and dashboard_id (from list_dashboards). Pass summary=true for ids, kinds and layout only, without the bindings.

get_datasheet

ChatGPT
Monthly values for a financial datasheet (CUSTOM numeric/formula and ACCOUNT KPI rows). Optionally pass datasheet_row_ids to return only those rows (ids from list_datasheet_rows). Formula cells are returned as formula / formulas ({Title} syntax). Includes number_format / cell_formats / decimals / cell_decimals when set (percent values are fractions: 0.5 = 50%). Only CUSTOM ids are bindable as datasheetRow. Requires company_id (from list_companies or the user).

get_default_budget

ChatGPT
Company default budget only. Response: { id: -3, name, type: 2, startDate, endDate } — use when the user asked for default budget (or actual vs default budget) and no other budgets or budget filter are needed. Pass budget_id: -3 and budget_type: 2 on statement requests. Do not bake name into generated HTML; at runtime widgets call /api/data/default-budget (or /api/data/budgets) for the current label. Requires company_id (from list_companies or the user).

get_default_forecast

ChatGPT
Company default forecast only. Response: { id: -4, name, type: 2, startDate, endDate } — use when the user asked for default forecast (or actual vs default forecast) and no other budgets or budget filter are needed. Pass budget_id: -4 and budget_type: 2 on statement requests. Do not bake name into generated HTML; at runtime widgets call /api/data/default-forecast (or /api/data/budgets) for the current label. Requires company_id (from list_companies or the user).

get_excel_sheet

ChatGPT
Monthly values for an Excel-uploaded spreadsheet in Reach (use sheet_id from list_excel_sheets). Same response shape as get_google_sheet. Optional sheet_row_ids limits to those from list_excel_sheet_rows. Requires company_id (from list_companies or the user).

get_excel_sheet_non_date_values

ChatGPT
Non-date column values for an Excel-uploaded sheet (same response shape as get_google_sheet_non_date_values). Use header ids from list_excel_sheets and row ids from list_excel_sheet_rows. Requires company_id (from list_companies or the user).

get_fiscal_year_start_month

ChatGPT
First calendar month of the company fiscal year. Response: fiscalYearStartMonth (1-12, January=1) and monthName (English). Use for fiscal-year labels or ranges when the user refers to FY rather than calendar years. At runtime, widgets may call /api/data/fiscal_year_start_month (same params as book_month) instead of baking this into HTML. Requires company_id (from list_companies or the user).

get_formula_capabilities

ChatGPT
Closed list of Excel functions valid in create_metric / update_metric formulas, each with a definition and syntax. Cell values use =FUNCTION(...) with A1 refs; CUSTOM row formulas use {Title} refs. Do not invent function names. Not views, styles, or Additional Settings — those are get_metric_capabilities.

get_google_sheet

ChatGPT
Monthly values for a Google-linked spreadsheet in Reach (use sheet_id from list_google_sheets). Optional sheet_row_ids limits to those from list_google_sheet_rows. Requires company_id (from list_companies or the user).

get_google_sheet_non_date_values

ChatGPT
Non-date column values for a Google-linked sheet: columns (metadata for requested header ids) and per-row values aligned to that order. Header ids are from each sheet’s non_date_headers in list_google_sheets. Same semantics as get_excel_sheet_non_date_values. Requires company_id (from list_companies or the user).

get_image_upload_status

ChatGPT
Find out whether the user has dropped their image on the page request_image_upload handed them, using the upload_id it returned. Returns state "waiting" (nothing has arrived yet), "complete" with the image — { id, name, url, width, height } — or "expired" once the link and its record have lapsed. Place a completed image by using its url as an image element href, and size the box from width and height so it is not stretched. Ask the user before checking again rather than polling: they have to find the file and drop it, which takes as long as it takes.

get_metric

ChatGPT
Read a dashboard or report tableKpi without writing. Pass company_id and table_id (create_metric return, IR kpiId, or statement page table_id from get_report_contents). Returns rows/columns (always: id, type, dates, range, selected; ACCOUNT rows also include item with companyId), persist cellStyles keyed by rows[].editorId / columns[].editorId, and formatting_rules decompiled to the same shape update_metric accepts (cells titles+YYYY-MM; statement sections+column). supports_statement_rules is true on locked statement tables; persisted statement rules are still returned when false. Use row_ids / YYYY-MM from this snapshot on update_metric. Not gated by report payload mode. Refuses missing/archived tables.

get_metric_capabilities

ChatGPT
Index of create_metric / update_metric / get_metric authoring: views, toolbar styles, Additional Settings extra (including extra.trendlines [left[], right[]] with seriesId + trendlineType), view_settings merge (see view_settings.invert for chart.inverted vs extra.invertRowsToColumns), row/column types, formatting_rules, and targeting. Closed vocabulary — do not invent keys. Trendlines are Additional Settings, not cell values; call get_metric before replacing extra.trendlines so the other axis is not wiped. update_metric requires lock_metric first (no same-user takeover) and unlock_metric after. Not the report page grammar — that is get_report_capabilities. A page tableKpi cannot carry formula, values, selected, column title/subtitle, formatting_rules, or view_settings; after compile, lock_metric and update_metric before print preview. create_metric can send them on its own call.

get_other

ChatGPT
Monthly Other / calendar values. Call list_other first and pass those ids as optional other_ids; with no other_ids, returns every available field. Requires start/end YYYY-MM (class, department, and budget filters do not apply). These are not statement rows and are not total_ids. Response matches ratio tools: months plus rows[] with node, values, bind, and format. Date values are YYYY-MM-DD strings; numeric values are integers. Requires company_id (from list_companies or the user).

get_profit_loss

ChatGPT
Profit & Loss (flat rows). Filter with account_ids, accountTotal_ids, and/or total_ids as in get_chart_of_accounts. With no id filters, returns all rows including GL lines and accountTotal/mapTotal rollups. Response: months (YYYY-MM[], once) and rows[] with node {value, name?}, values (numbers aligned to months, or null), and bind — the tableKpi item shape {itemType, identifier, companyId} — on bindable rows, so a row drops straight into a metric with no get_chart_of_accounts lookup. Requires company_id (from list_companies or the user).

get_ratios

ChatGPT
Monthly ratio values. Call list_ratios first and pass those ids as optional ratio_ids; with no ratio_ids, returns every available ratio. Same date filters as get_profit_loss (start/end YYYY-MM, optional class_id, department_id, budget_id, budget_type). Ratio ids are not total_ids for get_profit_loss. Response matches statement tools: months plus rows[] with node, values, bind, and format / help when the catalog has them. Percent values are fractions (0.42 = 42%). Requires company_id (from list_companies or the user).

get_report_capabilities

ChatGPT
Return an index of report-authoring capabilities for a company: the report toolbox vocabulary (element types, view types, tableKpi row/row-item/column kinds, liveText variable + dataType/comparisonType kinds), page geometry (paper sizes, orientations, viewBox per paper), the company theme palette (default cover/header/accent colors to it), ONE authoring grammar in authoring (tagged SVG by default; IR JSON when authoring_format=ir or IR-only mode — never both in one response), the company presentation settings (company_name, disclaimer, prepared_by, currency under data.company_settings), Other / calendar liveText fields (daysInMonth, bookMonth, etc. under data.other_fields with ready tableKpi_row binds; call get_other for values), the known-safe font families for report text (data.known_fonts — Montserrat is the default; pick another from this list when a different report font is requested) with the weights each offers (data.known_font_weights — pick fontWeight from the family's own list), and the company-scoped bindable handles (datasheets, google/excel sheets, classes, departments, budgets, plus book month + fiscal year start for date ranges) as id + name only. Fields the page schema cannot store (formula, values, selected, column title/subtitle, formatting_rules, extra.trendlines, styles, view_settings) are get_metric_capabilities, not this index. After the page compiles, lock_metric and update_metric before print preview. This is an INDEX of bindable handles, not raw accounting data — use the dedicated read tools (get_chart_of_accounts, get_totals, list_*_rows) to read account/total/row values.

get_report_contents

ChatGPT
Lift a CANVAS report for the modify-report loop (also works for temporary design-session reports before confirm_report). Call with company_id + report_id only to list pages (id, order, title, page_type, statement table_id) — no SVG bodies. Then call again with page_id (preferred) or page_order to fetch that page's schematic tagged SVG (data-reach-* attrs). Tiles are labeled boxes; tableKpi tags come from linked tables. One SVG = one page. Pass format="ir" to get the structured IR JSON instead of SVG — the same shape create_report_from_ir / update_report_from_ir accept, ~45% smaller, ideal for round-trip edits. Requires company_id from list_companies.

get_report_context

ChatGPT
Front-load the whole report-design profiling phase in ONE call. Returns: meta (company, book_month, fiscal_year_start_month, range, data_type); date_columns (resolved YYYY-MM-DD current/prior_period/prior_year/forecast, output_columns, default_basis, live_text_range, live_text_range2 — the last two are null for a budget report, as prior_period/prior_year already are; no client date math); statements (per requested statement: a populated flag, sections grouped by statement section with coverage-selected accounts, each account a ready-to-paste tableKpi row {type,item,title,originalTitle,dataType} + role-keyed values {current, prior|forecast, priorYear} + shareOfSection; each section carries accounts_total + truncated — when truncated, immaterial accounts were dropped, so call get_profit_loss for their values (binds are in accounts_index), plus a compact accounts_index of every active account {id,name,bind} for lookup); capabilities (toolbox + page geometry); handles (classes/departments, budgets only for budget, sheets on request); theme; images (logo + count always; full library only with include_images). Pass budget_type + budget_id together for Actuals vs Budget/Forecast. statements defaults to ["pl"]. include_grammar=false by default (fetch the grammar once via get_report_capabilities). Every account row already carries its bindable item AND title, so a metric binds with no get_chart_of_accounts lookup. WARNING: requesting more than one statement together with include_grammar:true in the same call risks exceeding the client output-size limit and can fail outright — never combine them; request one statement at a time and fetch the grammar via a separate get_report_capabilities call instead.

get_report_sample

ChatGPT
Fetch a curated sample report by id (from list_report_samples): the manifest summary plus its authored pages, each with page IR in the exact shape create_report_from_ir takes — submit it one page per call, in order. Pages are served bound to this company: every data item carries its company id, the cover image is its logo (dropped when it has none), report.themeColor is its primary theme color and the sample's accent colors are mapped slot-for-slot onto this company's palette, so total rows and variables (itemType "total") and the page colors need no change. Account-level rows and variables are role placeholders — identifier "role:<section>.<n>" (n-th largest account in that get_report_context section), "role:<section>.<word>" (an account of that kind), classIds "role:class.<n>" (n-th class) — that you re-bind by role using the bind-ready items get_report_context returns, or drop with their band when the section does not exist; the response lists them under bindings with instructions. Keep geometry, bands, chart views and narrative slots as they are; re-theme colors and retitle for the company. Pass page to fetch a single page when the full set is more than you need. A sample whose scan is "none" ships only a PDF: the response then carries pdf_url and the upload instructions instead of pages, and the PDF is transcribed at run time. Requires company_id from list_companies.

get_report_status

ChatGPT
Lightweight status for a CANVAS report: report last_modified, per-page last_modified (from page updatedAt), and current reportLock. Use after lock_report / update jobs and before follow-up edits to see if the user (or another save) changed the report. For a single-page follow-up, pass page_id — if that page last_modified matches your baseline, reuse the existing lifted SVG and do not call get_report_contents again for it.

get_theme_colors

ChatGPT
Hex color palette for charts: company chart colors from Reach, or the workspace default theme when the company uses workspace theme defaults. Returns colors (string[]) and source ("company" | "workspace"). Call once when building a widget and pass the array to Highcharts (Highcharts.setOptions({ colors }) or per-chart colors) so series match the org dashboard theme. Requires company_id (from list_companies or the user).

get_totals

ChatGPT
Total line keys from P&L, balance sheet, and cash flow layout rows: same as node.value with the trailing _{companyId} removed (e.g. total_income). Pass these as total_ids or use the same tokens without the total_ prefix where filters accept section tokens. Response: total_ids (sorted string[], deduped, casing preserved). Requires company_id (from list_companies or the user).

get_upload_instructions

ChatGPT
Return the rules for transcribing a user-attached PDF into the same shape a curated sample has: IR pages in the get_report_capabilities grammar plus a manifest. Call this on design_report's upload route, where no sample is picked and list_report_samples is therefore never called. A curated pick already carries the same text on the list_report_samples / get_report_sample result and does not need this. Mutates nothing. Requires company_id from list_companies.

list_budgets

ChatGPT
List budgets for the current company: company budgets (type 1), integration budgets (type 2), and budget groups (type 3). Each item has id, name, and type, including sentinel rows for the company default budget (id: -3, type: 2) and default forecast (id: -4, type: 2). Prefer get_default_budget or get_default_forecast when the user only asked for one default and no budget picker or other budgets are needed. When the user asked for defaults, pass -3 or -4 with type: 2 on statement requests—not the underlying budget id. Do not bake name values from tools into generated HTML; at runtime widgets call /api/data/budgets for current labels. Requires company_id (from list_companies or the user).

list_classes

ChatGPT
List accounting classes (segments) for the current company Requires company_id (from list_companies or the user).

list_companies

ChatGPT
List companies the authenticated user can access where AI is enabled (company + workspace, or demo). Call this first, then pass company_id on other tools. Consolidations include consolidation_companies (member ids for rows[].item.companyId and member_company_id on statement tools).

list_dashboards

ChatGPT
List a company's dashboards (id, name) so one can be picked by name. Company-owned active dashboards only — excludes presets, library templates and archived dashboards. Optional case-insensitive name filter; results are capped at 100, so pass a name to narrow a large set.

list_datasheet_rows

ChatGPT
List CUSTOM and ACCOUNT rows for one financial datasheet: id, title, type, plus item on ACCOUNT, formula when the row has Excel in customValues (decompiled to {Title}), and number_format / cell_formats / decimals / cell_decimals when set. Only CUSTOM ids are bindable as datasheetRow. Use sheet_id from list_datasheets. Requires company_id (from list_companies or the user).

list_datasheets

ChatGPT
List financial datasheets (KPI tables) with column months Requires company_id (from list_companies or the user).

list_departments

ChatGPT
List accounting departments for the current company Requires company_id (from list_companies or the user).

list_excel_sheet_rows

ChatGPT
List rows for one Excel-uploaded spreadsheet: each row id and title (label) in display order. Use sheet_id from list_excel_sheets. Requires company_id (from list_companies or the user).

list_excel_sheets

ChatGPT
List Excel-uploaded spreadsheets attached to the company along with their non-date headers. Each spreadsheet object has an id, name, and non_date_headers array of objects with id and name properties. Requires company_id (from list_companies or the user).

list_google_sheet_rows

ChatGPT
List rows for one Google-linked spreadsheet: each row id and title (label) in display order. Use sheet_id from list_google_sheets. Requires company_id (from list_companies or the user).

list_google_sheets

ChatGPT
List Google-linked spreadsheets attached to the company along with their non-date headers. Each spreadsheet object has an id, name, and non_date_headers array of objects with id and name properties. Requires company_id (from list_companies or the user).

list_image_library

ChatGPT
List the images available to place in this company’s report — the Live Logo plus the signed-in user’s uploaded Image Library assets — as one list, the same set the Reach image picker shows. Each item is { id, name, url, kind ("logo" | "image"), created_at, width, height }; the Live Logo is first. width/height are the image’s real pixel dimensions, measured server-side for any asset stored without them — use them to size the element proportionally rather than guessing. Place one by using its url as the href of an image element. The Live Logo is the exception — place it by reference (imageSource "reach", imageId "companyLogo", no href) so it keeps following company settings; its url here is for sizing and preview only. Optional case-insensitive name filter. Requires company_id (from list_companies); the logo is the company’s, the uploads are the signed-in user’s.

list_other

ChatGPT
Other / calendar data variables from the data picker (daysInMonth, netWorkingDays, numberOfMonths, date, bookMonth, lastSynced, startOfFiscalYear). Call this before get_other. Bind as tableKpi ACCOUNT rows or liveText info vars — see get_report_capabilities data.other_fields[].tableKpi_row. Not statement rows and not total_ids for get_profit_loss. Response: other[] with id, name, optional format, and bind ({itemType:"info", identifier, companyId} — omit integration). No values and no date range. Requires company_id (from list_companies or the user).

list_ratios

ChatGPT
Ratios / Metrics available for this company (gross profit margin, current ratio, EBIT, and any integration-specific ratios such as utilization). Call this before get_ratios. These are not statement rows and are not total_ids for get_profit_loss. Response: ratios[] with id (bare identifier), name, optional format (numberFormat: "percent" means values are fractions, 0.42 = 42%; optional decimals), optional help, and bind (tableKpi item {itemType:"total", identifier, companyId}). No values and no date range. Requires company_id (from list_companies or the user).

list_report_samples

ChatGPT
List the curated sample reports a new report can start from — call this only after design_report's earlier native pick (from scratch / start from a sample / upload my own) came back "start from a sample"; never call it before that choice is made. Each entry carries the manifest (title, tagline, audience, inferred range, orientation, paper, statements, statementPages, page list with roles / optional / requires), a content_page_count, and public pdf_url and preview_url for the USER to look at when choosing — never fetch the PDF yourself. On a client that renders MCP Apps this ALSO renders a gallery — one card per curated sample, with its thumbnail, tagline, page count and PDF link. The gallery is a visual reference and nothing more: the cards are not clickable and send NOTHING back, there is no card for "Upload my own" or "From scratch", and no message arrives from it — do not wait for one. YOU ask the question natively; the returned presentation field says what to say next — follow it, and never both show the gallery and restate the list. No page IR is returned here; call get_report_sample for a chosen sample. scan is "cached" when the sample ships pre-mapped IR pages and "none" when only the PDF exists — get_report_sample then returns pdf_url plus the upload instructions and the PDF is transcribed at run time like a user-attached one. Also returns upload_instructions: how to transcribe a user-attached PDF into the same shape. Requires company_id from list_companies.

list_reports

ChatGPT
List a company's editable CANVAS reports (id, name, description) so a report can be picked for the modify flow. Company-owned active reports only — excludes presets, library templates, and legacy (non-CANVAS) reports. Optional case-insensitive name filter; results are capped at 100, so pass a name to narrow a large set.

list_tags

ChatGPT
List TagInput tags for a company: id, label, the value the pill shows today, and the token to paste into a metric title. Paste token into view_settings title.text, subtitle.text, tableTitles[].text, or disclaimer.text; surrounding text is literal. Do not type value — that date is frozen and will not move with book month. Column header title and subtitle are plain text, not tags.

lock_dashboard

ChatGPT
Lock an EXISTING dashboard for Reach AI editing (modify_dashboard). Call after the user confirms which dashboard from list_dashboards and BEFORE get_dashboard_contents / update_dashboard_from_ir — the update is refused without this lock. While it is held, the dashboard is read-only in the Reach app for everyone, with a banner saying Reach AI is editing it. Fails, naming the holder, if another user holds an active Reach AI lock. Calling it again while you hold the lock refreshes it. Always call unlock_dashboard when the update job completes or you stop.

lock_datasheet

ChatGPT
Lock an existing company datasheet for Reach AI update_datasheet. Call AFTER the user confirms which datasheet and BEFORE update_datasheet. Fails if the Reach app editor is open — even for the same user; there is no takeover. Fails if another user or session holds an active lock. Always call unlock_datasheet when finished. Inactivity is 5 minutes without a write heartbeat.

lock_metric

ChatGPT
Lock an existing metric (dashboard or report tableKpi) for Reach AI update_metric. Call AFTER the user confirms which metric and BEFORE update_metric. Fails if the Reach app editor is open — even for the same user; there is no takeover. Fails if another user or session holds an active lock. Always call unlock_metric when finished. Inactivity is 5 minutes without a write heartbeat.

lock_report

ChatGPT
Lock an EXISTING confirmed CANVAS report for Reach AI editing (modify_report / create_metric). Call after the user confirms which report from list_reports and BEFORE get_report_contents / update_report_from_svg / create_metric. If the same user already has the report open in the Reach app editor, this takes the lock and that editor becomes read-only with live updates. Fails if another user holds an active app or AI lock. Always call unlock_report when the modify session is finished.

modify_dashboard

ChatGPT
Start an interactive session to CHANGE an existing Reach dashboard with the user. Use when the user wants to edit a dashboard they already have — NOT to build a new one (use design_dashboard) and NOT for a report (use modify_report). Requires company_id (from list_companies or the user). Mutates nothing — it returns the questions to ask and the step order to follow; the dashboard is found by name downstream with list_dashboards.

modify_report

ChatGPT
Start an interactive session to modify, edit, change, or update an EXISTING Reach report with the user. Use when the user wants to tweak a report they already have — NOT designing a new report (use design_report), and NOT the one-shot create_ai_report job. Requires company_id (from list_companies or the user). Mutates nothing — it returns the questions to ask and the step order to follow.

open_report_print_preview

ChatGPT
Open a report in Reach Print preview — a chrome-less, print-styled rendering of the report. Returns a time-limited URL. Works for temporary design-session reports before confirm_report — call this so the user can review before approving. This tool renders a UI widget (a clickable button) in the calling turn — if you are an orchestrator delegating report work to a subagent, do NOT let the subagent call this tool; have it stop short and return report_id/company_id to you, then call this yourself in the top-level conversation so the button actually renders for the user. After calling, render the report by opening the url in the built-in browser; if that is unavailable, open it in an external browser; only as a last resort, share the url as a link. An already-open preview tab auto-refreshes when update_report_from_svg completes — still call this for a URL, but do not ask the user to hard-refresh. Prefer this when the user wants to preview or print a report. Call list_companies first if you need a company_id; access requires company VIEW permission.

reorder_report_pages

ChatGPT
Reorder the pages of a report. Pass report_page_ids: the COMPLETE list of the report's page ids in the desired order — read the current pages (id + order) from get_report_contents first. The list must name every page exactly once; a partial or padded list is rejected. Requires company_id + report_id. Does not add or remove pages (use update_report_from_svg to add, delete_report_page to remove).

request_image_upload

ChatGPT
Get a short-lived link the user can drop an image file on, for an image that exists only on their own machine and has no URL. Put the page in front of them — open it for them if you can show them a browser — but do NOT operate it yourself, because the file is on their computer, not yours. They drop one PNG or JPEG on the page; you then call get_image_upload_status with the returned upload_id to find out what arrived, and place it by using its url as an image element href. The link lasts 15 minutes and works once. When the user instead names or links an image that is reachable on the web, use add_image_to_library and pass that URL — this tool is only for a file you cannot reach. For an image that does not exist at all, neither tool helps; say so rather than offering to make one. Requires company_id (from list_companies).

save_uploaded_image

ChatGPT
Put an image you have already PUT to a create_image_upload_url target into the user’s Image Library, using the upload_id that target returned. Returns { id, name, url, width, height }; place it by using the url as an image element href, and size the box from width and height so it is not stretched. Call this only after the PUT succeeded — if the bytes never arrived it refuses, and the target works once.

set_book_month

ChatGPT
Pin (fix) the company book month so live date columns, statement pages, and live sentences re-date to that period. This is a company-GLOBAL setting: it moves the book month for EVERY report and dashboard of the company, not just one report. Requires company_id and book_month (YYYY-MM or YYYY-MM-DD, normalized to the first of the month). After setting it, lift the report (get_report_contents) and review/rewrite its live text against the new-period figures. Use modify_report for the lift-and-rewrite flow.

unlock_dashboard

ChatGPT
Unlock a dashboard previously locked with lock_dashboard in the modify_dashboard flow, so it can be edited in the Reach app again. Call as soon as the update job completes (get_ai_job_status), or when you stop without submitting. Idempotent if already unlocked. Cannot unlock a lock held by another user.

unlock_datasheet

ChatGPT
Unlock a datasheet previously locked with lock_datasheet. Call when the update session is finished. Idempotent if already unlocked. Cannot unlock a lock held by the Reach app or another user.

unlock_metric

ChatGPT
Unlock a metric previously locked with lock_metric. Call when the update session is finished. Idempotent if already unlocked. Cannot unlock a lock held by the Reach app or another user.

unlock_report

ChatGPT
Unlock a CANVAS report previously locked with lock_report in the modify_report flow. Call when the modify session is finished (after open_report_print_preview / the user is done). Idempotent if already unlocked. Cannot unlock a lock held by the Reach app or another user.

update_dashboard_from_ir

ChatGPT
Apply changes to an EXISTING Reach dashboard in ONE call carrying the whole surface — a dashboard is a single grid, so there is no page chain. Requires the lock from lock_dashboard; refused without it. Read it first with get_dashboard_contents, change the tiles, and send them all back. Keep each tile’s id: a tile sent with its id keeps its metric and everything update_metric set on it, and a tile left out is REMOVED from the dashboard. Pass unsupported tiles back exactly as received: omitting a locked statement removes it, and omitting a live text, query table or AI widget frees the cells it holds so a later tile can land on top of it. last_modified is required and must come from the same read, so a concurrent edit in the Reach app is caught rather than overwritten. Tiles carry {x, y, w, h} in whole grid cells (max 60 tiles); nothing reflows, so close gaps yourself. Returns a job_id to poll with get_ai_job_status.

update_datasheet

ChatGPT
Edit a company datasheet (not Google/Excel, not a budget datasheet). Pass company_id and sheet_id from list_datasheets. Optional name; optional start/end as YYYY-MM (rebuilds DATE columns). Optional rows upsert mixed CUSTOM+ACCOUNT: omit id to create, pass id to update, id+delete to remove. replace_rows true means rows is the full mixed set (unlisted CUSTOM and ACCOUNT are deleted). CUSTOM values { "YYYY-MM": number } or formula "={Title}-{COGS}"; ACCOUNT { item_type, identifier }. Optional number_format / cell_formats / decimals / cell_decimals (percent values are fractions: 0.5 = 50%). Requires lock_datasheet first (no same-user takeover). Same snapshot return as create_datasheet. Caps: 200 CUSTOM + 200 ACCOUNT, 60 months.

update_metric

ChatGPT
Format a dashboard or report tableKpi. Pass company_id and table_id. styles = Formatting toolbar persist keys; omit leaves cellStyles. clear_styles clears the selection. view_settings deep-merges into viewSettings[current view] (TABLE extra: hideZero, zeroFormat, hideZeroRows/Cols, stripeRows, useThousands; plus titles/disclaimer. Charts: legend, yAxis, xAxis, series, extra). default_labels / default_abbreviate / default_hide_category_zero_rows are table flags. Targeting: no row_ids + styles → rowStyle on all data rows; row_ids → those rows; row_ids + columns → those YYYY-MM or editorId cells only. row_ids -2/-1 = column titles/subtitles; columns "-1" or row_title = row titles. apply defaults to row without columns, cells with columns. view_settings-only is valid. rows upsert ACCOUNT/CUSTOM/BLANK: omit id to create, id to update, id+delete to remove; replace_rows=true is the full set. CUSTOM formula ={Title}-{COGS} is row-wide. values (number or =Excel) key by YYYY-MM or editorId into customValues; ACCOUNT values only on custom columns. Excel functions: get_formula_capabilities. Cap 30. formatting_rules = value-based conditional formatting (not styles). cells: rows+columns+op+value; icon_color defaults to #333333. statement: sections+column on locked statement tables. replace_formatting_rules replaces the set. Cap 20. column_specs: date/forecast need startDate/endDate; custom has no dates and no live account pull — row totals use rows[].values =SUM(A1:L1) keyed by editorId. title on custom columns by default; date/forecast title only if the user asks. subtitle on any column. replace_columns is the full set. Cap 24. selected on rows and column_specs hides or shows; omit leaves the current value. view switches table|column|line|bar|pie|scatter|waterfall. yAxis/xAxis merge by index. Additional Settings (extra.trendlines, styles, plotLines): get_metric_capabilities. Inspect columns + view_settings with get_metric first. Requires lock_metric first (no same-user takeover). Refuses client portal.

update_report_from_ir

ChatGPT
Update an EXISTING CANVAS report from Reach IR JSON (schema: reach-svg-ir), ONE PAGE PER CALL. OVERWRITE a page: pass report_page_id + page. INSERT a NEW page: pass page and OMIT report_page_id — a fresh page is created and placed after insert_after_page_id (or at the end); the response returns its new report_page_id. First call: company_id, report_id, (report_page_id | insert_after_page_id), page, finalize=false → draft_id. Later pages: draft_id + page (+ report_page_id to overwrite). Last page: finalize=true (+ page) → job_id; or a follow-up with draft_id + finalize=true and no page to finalize pages already stored. Poll get_ai_job_status, then confirm_report. Same tableKpi rules as update_report_from_svg. At most 12 pages. Do NOT pass multiple pages in one call. On a retry reuse the returned draft_id + report_page_id so it overwrites rather than inserting a duplicate. To change the report page numbers (e.g. startIndex 0 after deleting the cover), set report.pageNumberSettings on the FIRST page (order 0) and resubmit it; it is merged over the current settings at compile. Settings on any other page are ignored.

Capabilities

InteractiveWrites

Example Prompts

Click any prompt to copy it.

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/reach-reporting"><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/reach-reporting)