MCP App Store
by KeenEthics
Dreamdata icon

Dreamdata MCP Server for Claude and ChatGPT

By Dreamdata.io ApS

Overview

Bring your complete GTM context to Claude. Every touchpoint is already organized by account and tied to revenue in Dreamdata’s account-based data model, so Claude answers from your complete GTM data without rebuilding context with every prompt. Get the same answer, no matter who’s asking. Claude uses Dreamdata’s semantic layer to define every metric and how to analyze it. So however people phrase their questions to Claude, your entire team lands on the same number. Verify any Claude answer. Whether it builds a new report or runs one you already have in Dreamdata, you can go to the report configurator to see the metrics, filters, date range, and attribution model behind it. You don't have to take our word for it, or Claude's.

How to connect

  1. Claude: Open the Dreamdata listing in the Claude connectors directory and choose Connect. Claude listing Dreamdata exposes 24 tools here.
  2. ChatGPT: Open the Dreamdata listing in the ChatGPT app directory and choose Connect. ChatGPT listing Dreamdata exposes 27 tools here.

Tools

create_dashboard

ChatGPT
Create a complete Analytics Hub dashboard — reports, layout and text — in one call. ONLY when the user asks for a dashboard. It writes to their Dreamdata account: one AI dashboard whose reports belong to it. Those reports do not appear in the account's report list, and are deleted with the dashboard. Workflow: 1. Plan the dashboard with the user first: the questions it answers, and one time range that every report shares (a dashboard has no time range of its own — each widget shows its report's). 2. Build and validate each report with the report_builder_ tools, ending with report_builder_run_config. Do NOT call report_builder_save_report for reports that belong to a dashboard: pass their configs in `reports`, keyed by a short local name. Existing saved reports can be used by `report_id`; the dashboard gets its own copy and the saved report is not changed. 3. Call create_dashboard once. The first call writes nothing: it validates everything, lists every problem at once, and returns a preview and a `confirmation_token`. Show the user the preview and ask whether to create the dashboard. Only after they clearly say yes, call again with identical arguments plus that token. Never invent a token or confirm on their behalf. `rows` lists the dashboard top to bottom; each row holds 1-6 widgets, laid out left to right (never pass positions — they are computed). Widget types: - metric: one number. Needs `report`/`report_id` and `metric_id`. - chart: a metric over time (report needs time.timeAggregation); `chart_display_type` line (default) or stackedBar. - breakdown_chart: a metric by segment (report needs segmentBy). - table: the report's summary table. - over_time_table: a metric over time, as a table. - text: `markdown` with headings, paragraphs, bold, italic*, lists. Metrics are 2 columns wide; the other widgets share the rest of the 12-column row, at least 4 columns each, so a row fits at most 3 charts, tables or text widgets. At most 50 widgets. Widgets are titled from their metric unless title is set. folder_id puts the dashboard in an existing Analytics Hub folder (its UUID from list_saved_reports, or its unique name). Leave it out unless the user asks for a folder.

describe_audience_config_schema

ChatGPT
Return the full JSON schema for a run_audience_query condition group. Call this when building a non-trivial audience — it documents every conditionType (property / event / stage / signal) and its fields, the operator and time-shortcut enums, and the nested filterBy / andThen structures. run_audience_query itself takes config loosely so this heavy schema is fetched only when you actually need it, not on every tool listing.

get_account_currency

ChatGPT
The currency the account's money amounts are in. Report results mark money columns unit: "currency" (or a metric's displaySymbol: "currency") but never say which currency: every amount is already in the account's own currency. Call this when an answer has to name that currency — writing an amount with a symbol, or saying what currency the figures are in. Never guess a symbol or code instead. Returns currency: {code, symbol, symbol_after_amount}, or currency: null when the account has none set — then give amounts without a symbol or code.

get_audience_definition

ChatGPT
Load a saved audience definition by id. Returns metadata including name, description, audienceType, folders, and a serialized config field. JSON-parse that config and pass the resulting list straight to run_audience_query as its config, to execute the audience and read its size off audiences.totalRowCount. Carry two of this definition's fields over as well, renaming them to the tool's snake_case arguments: audienceType becomes audience_type, and includeAnonymousCompanies becomes include_anonymous_companies. The builder's own bookkeeping (id, ui_value/ui_label, a materialized dateRange) does not need removing first; run_audience_query ignores it.

get_company_journey

ChatGPT
Fetch the full journey detail for a company. Fans out to two endpoints concurrently and merges the results: - GET /v1/journeys/companies/{company_id} → company summary - GET /v1/journeys/companies/{company_id}/sessions (with optional stageId) The company_id is the opaque id returned by search_companies. stage_id filters sessions to those that contributed to a specific stage instance. Call list_company_journey_stages(company_id) first to discover valid ids — stages are configured per account and vary by customer. The summary also includes a stages list for convenience; list_company_journey_stages returns the same data as a standalone call without the full journey overhead. Returns a merged dict with: - summary: trimmed company profile (identity, engagement, CRM links and properties filtered to non-empty values, plus stages list) - sessions: the session-endpoint payload (empty arrays dropped) with the raw per-touchpoint sessions list replaced by sessions_grouped: touchpoints sharing (channel, campaign, source, event, contactId) collapse into one group stored columnar — a flat timestamps array (one entry per touchpoint, in order; a bare date means exactly midnight UTC), a count, and sparse quantities / conversions_by_index / signalNames_by_index maps keyed by the string index into timestamps (present only where a touchpoint's value is non-default). When every touchpoint in a group shares the same signal set, that collapses to a single group-level signalNames list instead of the per-index map. This is lossless — no touchpoint is dropped — it only removes repeated dimension keys and empty/default values so a full history fits in context.

list_audience_filter_properties

ChatGPT
List the account's audience filter property catalog. Returns one group per property object (events, companies, contacts) plus one stages group per revenue model; a stages group's revenueModel field is the stage name used by stage conditions in run_audience_query (matched against the stage_name property). Signal properties are not available through this catalog. The events group leads with event, the key an event condition filters on to select which event; the event name itself goes in the condition's values. Call list_audience_filter_values with property="event" for the account's event names — any_event matches any of them. The rest of the group are event attributes (url_clean, session_channel, ...), usable either as an event condition's own property or in its filterBy. Pass property_object to narrow the result to a single object. The catalog is cached per account for a few minutes.

list_audience_filter_values

ChatGPT
List distinct paginated values for a specific audience property. For stages properties, pass the revenueModel of the stages group the property came from (see list_audience_filter_properties). To read the next slice, pass the response's pageToken back as page_token with the same property arguments, until a response has no pageToken. The token is opaque and expires with the results it pages through.

list_audiences

ChatGPT
List saved audiences visible to the authenticated user. Calls GET /v1/audiences/folders?viewAll=true and returns a trimmed catalog: folder metadata (id, name, count) plus one lightweight entry per audience. The large serialized config field is omitted — pass a returned id to get_audience_definition to load the full definition, or to run_audience_query after parsing that config. Newer APIs return every permanent audience when viewAll=true. Older APIs ignored that query param and only returned unfoldered rows, so when the base payload still looks root-only while folders report a non-zero count, this tool fetches each non-empty folder and merges (deduped by id).

list_company_journey_stages

ChatGPT
List the pipeline stages a company has reached. Stages are per-account: each Dreamdata customer configures their own pipeline (e.g. MQL, SQL, Opportunity, Closed Won). Use this to discover which stage IDs are valid for get_company_journey(stage_id=...). A stage name may appear multiple times with different timestamp values (e.g. the company hit "Opportunity" twice). Each occurrence has its own id and can be filtered independently. Returns a list of: - id: opaque stage instance id; pass as stage_id to get_company_journey - name: human-readable stage name (e.g. "MQL") - timestamp: ISO timestamp when this stage was reached - stageValue: revenue/value associated with reaching this stage - daysToValue: days from first touch to reaching this stage - firstTouchChannel: channel that originated the journey

list_my_accounts

ChatGPT
List the Dreamdata accounts the authenticated user can access. Use this to discover which value to pass as the optional slug argument accepted by every other tool. If the user has exactly one account, other tools select it automatically and slug can be omitted.

list_report_metric_drilldown_entities

ChatGPT
Return the entity list behind a saved report metric value (a drilldown). This is not the same as run_saved_report's output_type=contact_list: run_saved_report returns a report-defined view; this tool returns the entities (companies, contacts, deals, stages, or events) that contribute to a specific metric's value. For report-defined slices, use run_saved_report. On the initial call, provide report_id, output_type, and metric_id. output_type selects which list to return: company_list, contact_list, deal_list, stage_list, or event_list. Do not guess the pair: run the report first (run_saved_report with no output_type, or report_builder_run_config) and read its drilldown_metrics, which lists each drillable metric with the one output_type it supports. Optionally pass segment_by_values to filter by segment properties. The response includes next_page_token; pass it back as page_token to fetch subsequent pages (omit all other args when paginating).

list_saved_reports

ChatGPT
List all saved reports visible to the authenticated user. Calls GET /v1/content-list/view?viewAll=true and returns a trimmed catalog: each item carries just the fields needed to identify a report (id, name, description, folder, visibility, author, timestamps). Pass the returned id to the run_saved_report tool to run one. Dashboards and non-report items are filtered out.

report_builder_get_components

ChatGPT
Get the building blocks for a chosen analysis (step 3 — your source of truth). Returns: - sections: the report description, the metrics, the filter and segment properties (with descriptions, operators and sample values) and the config schema, each item's ids grouped under internal. The config schema names the attribution models this account can use. - metric_groups: every metric group in the account — {metric_group, label, metric_count}. Use it to pick a metric_group for a follow-up call; it is never scoped away. - stage_filter_properties: the ids that populate metrics.stageMetrics.<model>.filterBy, keyed by metric group — returned ONLY when metric_group names one stage group exactly. Otherwise you get stage_filter_properties_note instead; they are omitted because each stage model repeats the same properties under different hashed ids, which is what makes this response explode on accounts with many stage models. Build the report config using ONLY the metric and property IDs returned here — never invent IDs from labels. IDs are opaque hashes (e.g. p_y54efhqz2gmhgdf, m_zfp2zcnivsdzqvx) — copy them exactly as returned. Never construct or guess IDs; they are not derivable from labels or any pattern. metric_group scoping: pass an exact group id from metric_groups (e.g. "activityCountMetrics" or "stageMetrics.stage_model_abc") to cut the response down and to get that stage's filter properties. The "stageMetrics" family shorthand keeps every stage model's metrics, so it barely reduces size and never returns stage filter properties — reach for one exact stage group per call instead (repeat the call per stage if a report spans several).

report_builder_list_analysis_types

ChatGPT
List the analysis types a report can be built on. Each entry carries a summary and selection_guidance to help choose between neighbouring types. This is step 1 of the report-building flow; report_builder_start returns the same list plus the full workflow.

report_builder_list_property_values

ChatGPT
List the most common values for a filter/segment property (browse). Use when the user wants to see what's available ("what channels can I filter on?"). Returns the most frequent values. If the user already named a value, use report_builder_search_property_values instead. The result is a single page. When more values exist, totals.has_more says so (with totals.total). Do not try to page through them: tell the user roughly how many there are and search for what they are after.

report_builder_list_subjects

ChatGPT
List the analysis subjects available for an analysis type (step 2). The subject narrows what the analysis is about (e.g. which pipeline stage, or all contacts). Pick one, then fetch its details.

report_builder_resolve_calendar_period

ChatGPT
Turn a named calendar period into a report time filter (reads no account data). Use for a specific month ("May 2025" → year=2025, month=5), quarter ("Q1 2024" → year=2024, quarter=1), or whole year ("2023" → year=2023). Pass month OR quarter, never both. Returns a value shortcut or a frozen start/end pair to copy into config.time, plus an aggregation_hint. Never compute report dates yourself — always use this.

report_builder_resolve_rolling_window

ChatGPT
Turn a rolling window into a report time filter (reads no account data). Use for "last N days/weeks/months/years" (e.g. n=45, unit="days"). When the window matches a live shortcut exactly it returns a value token (put it in time.value); otherwise it returns a frozen start/end pair. Includes an aggregation_hint. Never compute report dates yourself — always use this.

report_builder_run_config

ChatGPT
Validate and run a report config; returns results or a fixable error (step 6). There is no separate validation step — this tool validates the config before running it. Omit output_type to get the full payload (and a supported_output_types field listing the slices this report can produce); or name a single slice (e.g. summary_table, time_series, company_list) to fetch just that view — call once without it first to learn which slices are supported. On an invalid config it raises with the API's structured validation errors in the message; read them, fix the config using the metric and property IDs from report_builder_get_components, and call this again. Results are paged. A single-slice response carries nextPageToken; the full payload carries nextPageTokens keyed by wire-shape name (summaryTable, timeSeries). When a token is non-null there are more rows: pass it back as page_token to fetch the next page, and omit every other argument — including config, which the token already encodes, so never re-send it just to page. A paginating call returns {rows, next_page_token, total_rows}; keep passing each next_page_token until it comes back null. Page size is fixed when the first token is issued and cannot be changed while paging. Money amounts (columns with unit: "currency", or metrics whose displaySymbol is "currency") are in the account's own currency, which the result does not name. To write one with a symbol or code, call get_account_currency first; never guess a currency.

report_builder_save_report

ChatGPT
Save a report the user configured (step 7 — ONLY on explicit user request). NEVER call this unless the user has explicitly asked to save the report — saving writes to their Dreamdata account (the report appears in the Analytics Hub for everyone on the account). Pass the exact config that report_builder_run_config last ran successfully — the server normalizes it to the shape the web app saves (metric display order, default ordering, explicit defaults, bare dates) so the report renders correctly in the UI configurator; do not add those presentation fields yourself. folder_id is optional. Prefer the folder UUID from list_saved_reports()["folders"][].id. A unique folder name is also accepted and resolved to that UUID. Do not invent folder ids. Confirmation protocol: the first call does NOT save — it validates the config, then returns a preview and a confirmation_token (an invalid config is rejected instead). Show the user the name, description, and folder (when set) and ask whether to save. Only after the user clearly says yes, call again with the same arguments plus that confirmation_token. Never invent or reuse a token, and never confirm on the user's behalf. (On clients that support elicitation the server asks the user directly instead, and no token is needed.)

report_builder_search_property_values

ChatGPT
Search a property's values by case-insensitive substring (validate). Use this before writing any user-named value into a filter (e.g. "paid social" → search="paid social") to confirm the value exists and catch spelling near-misses. Returns only matching values; the match runs over every value of the property, not just the most frequent ones. The result is a single page. When more values match than were returned, totals.has_more says so (with totals.total): ask the user for a more specific term rather than paging.

report_builder_start

ChatGPT
Begin building a Dreamdata report. Call this first. Returns the ordered report-building workflow plus the analysis types you can build on (each with selection guidance for choosing between them). Walk the steps in order — every subsequent tool's response tells you what to call next. Use this whenever a user wants to create, build, or set up a new report from scratch (to run an existing saved report, use run_saved_report instead).

run_audience_query

ChatGPT
Run an audience definition and return matching contacts or companies. config is a list of groups. Conditions inside a group are OR'd, while groups are AND'd. Each condition is an object with a conditionType (property, event, stage, or signal) plus camelCase structural fields (propertySource, propertyObject, …); the property value itself is an account-specific key returned by list_audience_filter_properties. Each condition's propertyType is resolved automatically from that catalog, so it can be omitted. Condition values are always strings, including numbers (e.g. ["0.5"]). For the full condition structure — every conditionType's fields, the operator and time-shortcut enums, and nested filterBy / andThen — call describe_audience_config_schema. Use list_audience_filter_properties to discover account-specific properties and list_audience_filter_values to enumerate valid values. Prefer the operators a property's allowedOperators lists: they are the ones that make sense for its type, though a few others are also accepted. Date and datetime properties are the strict case — only before/after, each needing one timestamp value, with no null check. Event conditions select the event with property="event" and the event name in values (any_event matches any event); use list_audience_filter_values with property="event" to list them. An event condition may instead filter directly on an event attribute such as url_clean. eventType defaults to at_least and count to "1", so "performed this event" needs neither. Set count for a specific number, and give it explicitly with eventType="at_most" or "only" — neither has a safe default. Use eventType="only" with count="0" for "never performed". count may be written as a number or a string. A chain step in andThen takes no count, and a signal condition takes no eventType. In an andThen chain, order is taken from the list itself — the steps run in the order given, and you never set a position field. A config from get_audience_definition can be passed as-is: the query builder's bookkeeping (id, ui_value/ui_label, filterBy: false, and a dateRange materialized beside a time shortcut) is ignored. A saved custom_period window is the one exception — its dateRange is not part of this contract and is reported rather than silently reinterpreted. sort_by takes a result column alias, not a property key. For companies: companyId, name, website, country, industry, accountOwnerName, numberOfEmployees, contacts, channels, sessions, lastSeen, linkedinUrl, isNew, engagementScore, engagementScoreTrend, activityTrend, sourceSystem. For contacts: contactId, email, name, title, userCountry, lastSeen, companies, sourceSystem. Stage conditions match on the stage_name property; valid stage names are the revenueModel values of the stages groups returned by list_audience_filter_properties. Pagination: audiences.totalRowCount is the full match count, of which each response returns one page_size slice. To read the next slice, pass the response's audiences.pageToken back as page_token together with the same audience_type and config, and keep going until a response has no pageToken. Treat the token as opaque — it also identifies the query results it pages through, so it cannot be edited, split, swapped with a token from another tool, or reused once those results expire (about 24 hours). Examples: - Companies in Fintech industry: [{"conditions": [{"conditionType": "property", "propertySource": "dreamdata", "propertyObject": "companies", "property": "industry", "operator": "equals", "values": ["Fintech"]}]}] - Companies with a 30-day engagement score above 0.5: [{"conditions": [{"conditionType": "property", "propertySource": "dreamdata", "propertyObject": "companies", "property": "engagement_score_30_days", "operator": "greater_than", "values": ["0.5"]}]}] - Contacts who performed page_view at least 3 times in the last 30 days: [{"conditions": [{"conditionType": "event", "propertySource": "dreamdata", "propertyObj…

run_saved_report

ChatGPT
Generate a saved report on behalf of the authenticated user. output_type selects a report-defined view (the metric cards, summary table, or list slices the report itself produces). Omit it to fetch the full payload — that response includes a supported_output_types field listing exactly which slice names this report accepts, so call it first when you don't yet know what's available. It also includes drilldown_metrics: the (metric_id, output_type) pairs list_report_metric_drilldown_entities will accept for this report. Results are paged. A single-slice response carries nextPageToken; the full payload carries nextPageTokens keyed by wire-shape name (summaryTable, timeSeries). When a token is non-null there are more rows: pass it back as page_token to fetch the next page, and omit every other argument — a paginating call resolves nothing from the report and returns {rows, next_page_token, total_rows}. Keep passing each response's next_page_token until it comes back null. Page size is fixed when the first token is issued and cannot be changed while paging. Money amounts (columns with unit: "currency", or metrics whose displaySymbol is "currency") are in the account's own currency, which the result does not name. To write one with a symbol or code, call get_account_currency first; never guess a currency. For entities behind a specific metric value (a drilldown), use list_report_metric_drilldown_entities instead — its output_type names look the same (contact_list, etc.) but mean "entities behind this metric," not "report-defined view."

run_saved_report_for_date_range

ChatGPT
Run a saved report over a custom date range, without saving any change. Loads the saved report and overrides ONLY its time window with the given absolute range — every other setting (filters, metrics, segmentation, time aggregation) is inherited from the saved report unchanged. The report is run and the full result payload is returned (metrics, summaryTable, timeSeries, nextPageTokens) plus a supported_output_types field, unless output_type narrows it to a single slice. Nothing is persisted. start_date and end_date are YYYY-MM-DD and both inclusive; pass the same date as both for a single day. The end must be on or after the start, on or before today, and the start no earlier than 2006-01-01. output_type works exactly as it does on run_saved_report: omit it for the full payload (which carries supported_output_types), or name one slice (e.g. summary_table, time_series, company_list) to fetch just that view. Results are paged. nextPageTokens is keyed by wire-shape name (summaryTable, timeSeries); when one of those is non-null there are more rows for that slice. Pass it back as page_token — omitting every other argument — to fetch the next page, which returns {rows, next_page_token, total_rows}. Keep passing each response's next_page_token until it comes back null. Page size is fixed when the first token is issued and cannot be changed while paging. The date range is baked into the token, so do not re-send it when paginating. Money amounts (columns with unit: "currency", or metrics whose displaySymbol is "currency") are in the account's own currency, which the result does not name. To write one with a symbol or code, call get_account_currency first; never guess a currency. To run a report over its own saved date range, use run_saved_report instead.

save_audience

ChatGPT
Save an audience the user built (ONLY on explicit user request). NEVER call this unless the user has explicitly asked to save the audience — saving writes to their Dreamdata account, where the audience appears in the Audience Hub for everyone on the account and can be synced to ad platforms. Pass the same audience_type, config and include_anonymous_companies that run_audience_query last ran successfully; the server adds the fields the Audience Hub editor needs, so do not add ids or ui_value yourself. This always creates a new audience; it never changes an existing one. name is 1-65 characters and must not match an existing audience's name; description is at most 512 characters. folder_id is optional: pass an audience folder id from list_audiences()["folders"][].id, or a unique folder name. Report folders are separate and are not accepted. Confirmation protocol: the first call does NOT save — it validates the config by running it, then returns a preview (including the audience's current size) and a confirmation_token; a config that cannot run is rejected instead. Show the user the name, description, audience type and folder (when set) and ask whether to save. Only after the user clearly says yes, call again with identical arguments plus that confirmation_token. Never invent or reuse a token, and never confirm on the user's behalf.

search_companies

ChatGPT
Search the Companies view by name and return matching journey records. Calls GET /v1/journeys/companies/paginated and returns a trimmed list of company journey records matching the search query. Each item includes core identity fields, engagement metrics, a 30-day activity trend, and CRM source links. Pass the returned id to the get_company_journey tool to fetch the full journey detail.

describe_audience_config_schema

Claude

get_audience_definition

Claude

get_company_journey

Claude

list_audience_filter_properties

Claude

list_audience_filter_values

Claude

list_audiences

Claude

list_company_journey_stages

Claude

list_my_accounts

Claude

list_report_metric_drilldown_entities

Claude

list_saved_reports

Claude

report_builder_get_components

Claude

report_builder_list_analysis_types

Claude

report_builder_list_property_values

Claude

report_builder_list_subjects

Claude

report_builder_resolve_calendar_period

Claude

report_builder_resolve_rolling_window

Claude

report_builder_run_config

Claude

report_builder_save_report

Claude

report_builder_search_property_values

Claude

report_builder_start

Claude

run_audience_query

Claude

run_saved_report

Claude

run_saved_report_for_date_range

Claude

search_companies

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/dreamdata"><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/dreamdata)