Dreamdata

View saved version →

---
name: dreamdata-activation
description: Find companies or contacts using Dreamdata audience criteria, explore saved audiences, and create and save audiences for activation. Excludes reports, dashboards, and company journey analysis; use dreamdata-analytics for those workflows.
---

# Dreamdata activation

Use the connected Dreamdata MCP tools. Tool names below are suffixes; their host namespace can vary. If the connection is unavailable, follow the bundled [setup skill](../setup/SKILL.md) to authenticate and verify access.

Use this skill for audience discovery, querying, and saving. Use the [analytics skill](../dreamdata-analytics/SKILL.md) for reports, dashboards, metric drilldowns, and company journeys. For mixed requests, use each skill for its relevant portion.

## Account and audience context

Use `list_my_accounts` when the intended account is unknown. Select the account from the user's context; ask if several remain plausible. Keep its slug consistent. One-account users can omit the slug.

Make the audience type, filters, time windows, and account clear. Distinguish the full match count from a partial page of entities. Link source URLs returned by the tools; do not invent links.

## Query audiences
For a saved audience, use `list_audiences`, load it with `get_audience_definition`, parse its serialized config, and execute with `run_audience_query`.

For a new audience query, discover account-specific properties and valid values with `list_audience_filter_properties` and `list_audience_filter_values`. Read `describe_audience_config_schema` before composing non-trivial conditions. Conditions within a group are OR; groups are AND. Condition values are strings, including numeric values. Stage names come from the catalog's revenueModel values. The catalog does not supply signal properties; do not invent them.

Check each property's type before translating the user's wording into a condition. Size properties such as employee count and annual revenue are usually text properties holding range labels (for example `251 - 1k` or `> 100k`), so `greater_than` and `less_than` are not available. For a threshold such as "more than 1,000 employees", list the property's values, read each label's range, and add one equals condition per band that lies entirely above the threshold (conditions in a group are OR). If the threshold falls inside a band, say so and ask whether to include that band rather than guessing. Tell the user which bands you used. Confirm other named values, such as countries, with `list_audience_filter_values` and its `search_query` before using them.

Audience pagination differs from report pagination: pass `audiences.pageToken` with the same audience type, config and slug. `list_audience_filter_values` pages the same way: repeat the property arguments and slug with the token. Use `audiences.totalRowCount` for the full match count, and disclose when listed entities cover only a page. Running a query does not imply that an audience has been saved or exported.

## Save and activate

When the user requests saving an audience and save_audience is exposed, follow its current schema and confirmation flow. Show the audience name, type, folder, and size before confirmation. The optional audiences:write grant is needed. Saving creates a new audience; do not claim it overwrites an existing audience or activates a destination sync. Users can subsequently configure syncing in Dreamdata's Audience Hub. If the tool is absent, explain that limitation rather than inventing an API.

View saved version →

---
name: dreamdata-analytics
description: Analyze and build Dreamdata reports and dashboards, inspect company journeys, and drill into report metrics. Excludes audience discovery, querying, and saving; use dreamdata-activation for those workflows.
---

# Dreamdata analytics

Use the connected Dreamdata MCP tools. Tool names below are suffixes; their host namespace can vary. If the connection is unavailable, follow the bundled [setup skill](../setup/SKILL.md) to help the user authenticate and verify access.

Audience workflows belong to the separate [activation skill](../dreamdata-activation/SKILL.md). For mixed requests, use each skill for its relevant portion.

## Account and analysis context

Use `list_my_accounts` when the intended account is unknown. Select the account from the user's context; ask if several accounts remain plausible. Keep its slug consistent throughout the analysis. One-account users can omit the slug.

Make the account, reporting period, filters, metric definitions, and attribution model clear where they affect the answer. Preserve the user's chosen scope. Distinguish attributed revenue from causal impact, missing data from zero, and a partial result page from the complete population. Link returned source URLs when available; do not invent report links.

## Saved reports

Discover report IDs with `list_saved_reports`; resolve ambiguous names before running. Use `run_saved_report` for the saved period or `run_saved_report_for_date_range` for a requested date override, which does not persist changes. Omit `output_type` initially to discover supported slices.

For entities contributing to a specific metric, use `list_report_metric_drilldown_entities`. A report-defined company/contact view is not the same as a metric drilldown.

Report responses may contain `nextPageTokens` by slice or `nextPageToken`. Continue a slice by passing its token as `page_token` together with the same `slug` as the call that returned it, and no other arguments; later responses use `next_page_token`. Without the slug, users with several accounts get an account-selection error instead of the next page. Fetch enough pages to support the requested conclusion and disclose any incomplete coverage. Do not compute population totals or rankings from an incomplete slice.

## New reports

Start with `report_builder_start` and follow its ordered workflow. Discover the analysis type, subject, and components using the returned tools. Use only returned metric, property, subject, and attribution-model IDs. Request an exact metric group when stage-specific filter properties are needed.

Pipeline outcomes are the account's own stages, listed by `report_builder_list_subjects`; names vary between accounts (one has Closed Won, another NewBiz). Map the user's outcome to one of those stages before choosing metrics. Loose terms such as "new companies", "new customers", "leads" or "deals" don't name a metric: if exactly one stage clearly matches, use it and say which; if several could (for example Closed Won and NewBiz), or none does, ask which stage the user means. Never substitute a different metric silently, and if no report offers what was asked, say so and offer the closest stage-based alternative.

Validate user-named filter values with `report_builder_search_property_values`; use `report_builder_list_property_values` for browsing. Resolve named calendar periods and rolling windows using the corresponding `report_builder_resolve_calendar_period` or `report_builder_resolve_rolling_window` tool. Use the returned time filter instead of hand-calculating those windows.

Run with `report_builder_run_config`, initially omitting `output_type` to discover supported slices. Correct structured validation errors using the discovered components. If the same unresolved error repeats, report the limitation instead of looping. Running validates the config but does not save it.

Only use `report_builder_save_report` when the user explicitly requests saving. Pass the exact config last run successfully. Follow the server's confirmation protocol: present the preview's name, description, and folder; obtain the user's confirmation; then submit the same arguments with the returned confirmation token. If the server handles confirmation through elicitation, let it do so. Never invent a token or confirm for the user. Saved reports are visible to others on the account.

## Company journeys

Use `search_companies` to resolve a company and its opaque ID. For stage-specific analysis, discover stage instances with `list_company_journey_stages`; repeated stage names can have different timestamps and IDs. Fetch the journey with `get_company_journey` using the selected instance where relevant.

The journey's `sessions_grouped` is compressed, not sampled. A group has a timestamps array and a touchpoint count; sparse quantity/conversion/signal maps refer to timestamp indices. Do not count groups as individual touches. A bare date represents midnight UTC. Describe observed touchpoints and stage timing without implying that sequence alone proves causation.

## Currency
When get_account_currency is available, use it before attaching currency symbols to report amounts. Report values use the account currency; do not infer it from the user's location. If currency cannot be established, show amounts without an assumed symbol and explain the limitation.

## Dashboards

When create_dashboard is exposed and the user requests a dashboard, follow its current schema and preview/confirmation flow. It requires the optional dashboards:write grant. Show the proposed layout and widgets before saving. Use the live tool descriptions as the authority for these operations.

View saved version →

---
name: setup
description: Connect Dreamdata and verify account access after installing the Dreamdata plugin, or troubleshoot initial authentication and account selection.
---

# Set up Dreamdata

The plugin declares the shared MCP endpoint https://mcp.dreamdata.io/mcp in its root mcp.json. Users authenticate with their own Dreamdata login through OAuth 2.1 with SHA-256 PKCE. No existing custom connection or account-specific endpoint is required by this configuration.

## Connect and verify

If Dreamdata tools are available, call list_my_accounts to verify access. A successful response establishes connectivity; an installed plugin alone does not. If authentication is requested, direct the user through the host's Connect or sign-in flow. Never request passwords, tokens, or client secrets in chat or store them in plugin files.

If tools are unavailable, ask the user to enable the installed Dreamdata plugin and complete its MCP sign-in flow, then use a fresh chat if needed. Do not claim to create a connection without a supported host operation. Avoid creating a second custom connection when the bundled server can be authenticated.

For manual ChatGPT connection setup when the host cannot load the bundled server, follow the official Dreamdata instructions: use https://mcp.dreamdata.io/mcp, and clear OIDC enabled under Advanced OAuth settings > OpenID support before authenticating. Explain this as a fallback rather than an installation requirement.

Reading Dreamdata resources is required. Saving reports, audiences, and dashboards are separate optional grants. Explain these options based on the user's intended work; do not require write permissions for reading. OAuth consent does not authorize individual save operations. Let the user complete consent themselves.

Once connected, select the sole account automatically. If several accounts are available, use the user's stated account or ask them to choose. Do not store account selections in shared package files. Offer a relevant first task: analyze a saved report, inspect a company journey, or find companies for an audience. For reports, dashboards, and company journeys, use the [analytics skill](../dreamdata-analytics/SKILL.md). For audience discovery, querying, and saving, use the [activation skill](../dreamdata-activation/SKILL.md). Perform the task they choose using the available Dreamdata tools.

## Connection issues

- 401: reconnect to renew authentication.
- 403: verify the account against list_my_accounts; do not assume a missing scope is the cause.
- Missing save permission: explain the required optional grant. Dreamdata documents disconnecting and reconnecting to obtain newly available permissions; explicit-scope clients also need the scope added to their configuration. Do not disconnect automatically.
- Client not recognized: Dreamdata allowlists OAuth clients; direct the user to mcp-feedback@dreamdata.io.
- Server not enabled for the user: direct the user to their Dreamdata account owner or mcp-feedback@dreamdata.io.
- 502: report a Dreamdata service error, not a local configuration problem.

Reference: https://developer.dreamdata.io/mcp/mcp-server/