Minds: Synthetic Market Research Panels MCP Server

Creates a new Study workspace from existing Audiences or inline Audience configurations; it does not ask questions or run research. Follow-up research in an existing Study does not require another Study. This operation ALWAYS creates: a matching name never attaches to an existing Study, and only a byte-identical repeat of the same call (same name, Audience IDs, inline configs, and sharing flag) replays the Study the first call made, reported as replayed: true. Settle the complete Audience set before calling: no advertised tool adds an Audience to an existing Study, and Study deletion is unavailable in this app, so calling again with one extra Audience leaves a permanent duplicate. List the existing Studies first when the one you want may already exist. Every questionnaire, survey, battery, section, cohesive question set, or request with two or more known questions belongs in one planned and confirmed multi-question block inside that Study. Never submit such a known set as separate direct questions. A direct question is appropriate only when exactly one standalone question is known or the next question depends on earlier results. Composite creation is atomic and rolls back partial Audience failures. Studies are private by default; enabling link sharing also publishes the attached Audiences and Minds.

Returns durable progress or results for one confirmed multi-question run inside a Study, including its immutable confirmed plan, server-prepared execution plan, respondent-visible question audit, question progress, response artifacts, deterministic method calculations, and stage readiness. Progress is question-level (settled/total). Completed response artifacts contain every participating Mind’s answer and grow with respondent count; internal generation and language-review diagnostics are omitted.

Applicability: use only when the user asks to find, inspect, or check a Study or its export. For a deletion-only request, do not call this or another lookup tool: Study deletion is unavailable in this app. Explain the limitation and direct the user to the Minds workspace at https://getminds.ai/ or authorized support without first retrieving the Study. Returns the current Study state, progress for in-flight questions, completed per-Audience results, linked Minds, Study links, and the status of a specific asynchronous export job when export identifiers are supplied. Values can be numeric answers or classified summary labels; message fields contain the original Mind responses when available. locale is the Study display locale, not a guarantee of the language of every response. An in-flight question reports "0 Minds answered so far" and stays at zero for its whole run — partial per-Mind progress is not persisted — then jumps straight to the finished table, so zero is not evidence that the run is stuck. Pass questionId to follow one question: the response then covers only that question. Without it the response covers every question the Study has run, with the full per-Mind answer table in structuredContent, and grows with questions x Audiences x Minds x answer length. To follow a confirmed multi-question run, poll the multi-question run status operation instead, which returns question-level counters only. In the final reply, include a clickable link to the returned workspaceUrl, preserving the URL verbatim even when a Study widget is displayed. This is the authenticated owner's workspace link, not a public share link. Use a returned sharedStudyUrl only for public or recipient handoff.

Returns or refreshes the semantic summary for a Study as Markdown plus flexible evidence blocks. Website, image, and video analyses retain heatmap-compatible block metadata.

Lists durable unfinished study drafts, or returns the complete saved planning state for one exact draft ID. Draft records are distinct from running or completed studies.

List your own and team-shared Advanced Study templates, most used first, or read one exact template including its revision, research method, questions, response settings and question attachments. configuration.methodId names the registered research method the Study runs under. configuration.questions is a fixed, pre-registered instrument — exact wording, order and per-item response contracts — ready to be transcribed into a question plan exactly as saved, without being re-planned. Saved templates commonly cover pricing and willingness to pay, concept and message tests, A/B comparisons, website reviews, video ad analysis, usability tests and market exploration.

Executes the exact stored multi-question draft revision inside the target Study after explicit confirmation. One execution submits the ENTIRE draft—all named modules and all questions—as one durable run. Never execute each known question separately and never create one run per module. This operation is valid only after the person explicitly confirms the exact draft revision. The server reloads and validates the revision, method availability, runner configuration, reviewed capabilities, optional advanced-method consent, and idempotency key before queuing durable research. MCP cannot upload or read a local file:// path: attach the file through the client and use its signed HTTPS URL, upload it to Minds first, or use another fetchable HTTPS URL. Before invoking any Mind or using Study quota, the server verifies that required respondent-visible source material is readable and refuses the entire run if it is not. Answer carryover: every questionnaire item is an independent measurement. A confirmed run carries no answer history — a Mind answering item 12 has not seen its own answers to items 1-11, and items fan out in parallel — and nothing on this run can change that (historyMode is settable only on a single standalone question, and defaults to none there too). So cross-item order, fatigue, and carryover effects do not arise, and rotation or least-fill designs that exist only to control them buy nothing here; conversely, any item that must build on an earlier one has to restate what it needs in its own text or stimulus, because a back-reference resolves to nothing.

Read or start a question asset heatmap, with the same behavior as Minds UI. For a specific video or image pass assetKey: its saved upload path (chat/...) or normalized URL. Only assets assigned to that question can be analyzed. GET returns assetHeatmaps keyed by asset identity; start with assetKey reuses completed analysis for that asset, while start without assetKey can rerun analysis. Website analysis visits the assigned public URL. Starting analysis uses one response per Mind and requires Premium. Selecting a different video does not change the question results.

Exports an Audience brief through the same unified branded renderer used by the web app. Supports Markdown, PDF, DOCX, and PPTX. Binary artifacts are returned as base64.

Exports a completed website heatmap from a Study result, identified by the message ID reported with the completed result. Returns the same ZIP archive as the web app, including its unified-renderer PDF report, Markdown, images, and metadata.

Generates a branded profile for one existing Mind, identified by exact ID or fuzzy name. Markdown is returned inline by default; PDF, DOCX, and PPTX artifacts are returned as base64 with a workspace link.

Starts an asynchronous export of Study results and returns an export job ID. Supports executive briefs and full reports in PDF, DOCX, PPTX, or Markdown, plus raw data in CSV, XLS, or SPSS SAV.

Submits exactly one respondent-visible standalone question to one existing Audience, identified by exact ID or fuzzy name. It creates a private Study for that Audience, starts asynchronous responses from its Minds, and returns the Study identifier and links. Automatic classification may reformulate the question; this operation does not promise verbatim wording. Locked respondent wording and response formats require a reviewed and confirmed Study plan. Never use this operation for a questionnaire, battery, section, cohesive question set, or any request with two or more known questions; create a Study and plan the complete multi-question block instead. MCP cannot read or upload a local file:// path. Use a fetchable HTTP(S) URL, a signed URL supplied by the client for the attached file, or an existing Minds workspace upload URL/path. Study tools import external file URLs into durable Minds storage before saving or running. A workspace upload reused as a Study asset must sit under chat/<userId>/: request the signed upload with folder "chat" and pass that storage path or its /api/uploads/chat/ URL. A temp/, portfolio/, or context/ path — and any /api/uploads/file-access/ URL — is refused as not belonging to the workspace owner and is never re-fetched; pass a plain external URL instead. The Study refuses to start if Minds cannot read the asset. In the final reply, include a clickable link to the returned workspaceUrl, preserving the URL verbatim even when a Study widget is displayed. This is the authenticated owner's workspace link, not a public share link. Use a returned sharedStudyUrl only for public or recipient handoff.

Submits exactly ONE respondent-visible question in an existing Study. Applicability: one standalone question, or one adaptive follow-up whose wording could not be known before earlier results. Exclusion: never use this operation for a questionnaire, survey, battery, section, cohesive question set, or any request containing two or more known questions—even when every question targets the same Study. The complete known set belongs in one planned and confirmed multi-question block inside the Study and must be submitted once, not question by question. Follow-up questions remain within the existing Study; this operation does not create or enumerate Studies. The full question value may reach respondents and is not planner-only metadata. Scale, categorical, and qualitative questions are classified automatically, and the response includes status and workspace links. Automatic classification may reformulate the question; this operation does not promise verbatim wording. Locked respondent wording and response formats require a reviewed and confirmed Study plan. The question is classified before it is queued, so the returned status is authoritative: queued or running means it was submitted to respondents, while planning_required means it was declined as an unrefined research objective and nothing was submitted. A planning_required response carries the original request plus a proposed headline and respondent questions for study planning. MCP cannot read or upload a local file:// path. Use a fetchable HTTP(S) URL, a signed URL supplied by the client for the attached file, or an existing Minds workspace upload URL/path. Study tools import external file URLs into durable Minds storage before saving or running. A workspace upload reused as a Study asset must sit under chat/<userId>/: request the signed upload with folder "chat" and pass that storage path or its /api/uploads/chat/ URL. A temp/, portfolio/, or context/ path — and any /api/uploads/file-access/ URL — is refused as not belonging to the workspace owner and is never re-fetched; pass a plain external URL instead. The Study refuses to start if Minds cannot read the asset. In the final reply, include a clickable link to the returned workspaceUrl, preserving the URL verbatim even when a Study widget is displayed. This is the authenticated owner's workspace link, not a public share link. Use a returned sharedStudyUrl only for public or recipient handoff.

Supports operationId-only requests to read a previously accepted preview or creation job and retrieve its result without creating another Audience. Creates a grounded Audience of synthetic Minds from a population or audience brief. It can combine authoritative web research, supplied sources, research files, and reviewed spreadsheet distributions; persists provenance and allocation audits; supports balanced, segment-coverage, and benchmark-depth sizing; generates the same member portraits and Audience cover as in-app creation; individually trains every member, including large reviewed-dataset cohorts, in the background (wait for every member to report readyToChat before running a Study); is idempotent for identical inputs; and keeps the Audience private unless link sharing is enabled.

**Call this before creating an Audience whenever a size is in play** — the user named a number, asked for "as many as possible", or you are choosing a creation mode. Returns the ceilings that apply to the authenticated connector account: the per-Audience plan cap including any configured team allowance, the custom-size maximum, and per-mode ceilings. The response identifies scope.userId, scope.teamId, allowance.source and checkedAt. Match that account/team to the intended workspace before claiming its limit; another connector or browser login can have a different allowance. These are sizes for each Audience, not total workspace capacity, remaining credits or a reservation of capacity. Read the two mode ceilings carefully, they are different things: `automaticSizingCeiling` bounds the size the server picks when you send no memberCount, while `explicitCountCeiling` bounds a size you state. Only "balanced" has the latter — exceed it and creation is refused with MODE_CAP. In the deeper modes a stated size is bounded by `memberCap` alone. Prefer calling this over guessing: being refused mid-creation is expensive, because the build has already been enqueued by then.

Lists the authenticated user's Audiences, most recently updated first, one page at a time (limit, default 20, and offset; nextOffset continues), with Mind counts, sharing state, and workspace or shared links. includeMinds adds each Audience's member Minds. searchQuery returns the best fuzzy name match instead of a page. In the final reply, include clickable links to each returned Audience workspaceUrl and the top-level workspaceUrl, preserving the URLs verbatim. These are authenticated workspace links, not public share links. Use sharedAudienceUrl only for public or recipient handoff.

Imports supplied UTF-8 text, Markdown, CSV and JSON research files into account-owned storage. Accepts file contents rather than local paths. Identical file imports are safe to retry. Optional caller-reviewed grounding JSON binds distributions to existingFiles followed by files in sourceIdx order, returning a normalized snapshot and checksum for audience preview and creation. This operation creates no Audience or Minds and performs no web search or independent verification of supplied percentages.

Save, explicitly update, delete or use a Custom research template. Templates are private by default. Owners can set update.isSharedWithTeam to share with their current team or turn sharing off. Teammates can read and use shared templates; only owners can update, delete or change sharing. Use creates an independent editable draft in the Minds web app and never starts research; that draft is finished in the app rather than over MCP. The stored configuration.questions of a template can instead be transcribed into a question plan here as a fixed instrument. Save exact questions, response settings, question attachments and the research method in configuration.methodId. A calculator-backed method that designs its own tasks at execution (Van Westendorp, Gabor-Granger, MaxDiff, conjoint, Kano, NPS, top-box, key drivers, TURF) writes those questions when the Study runs, and the saved questions are asked alongside them. Context and Audiences are always entered fresh. Provide save.requestId for retry-safe creation. Updates and use require the current revision. Use also requires a fresh requestId UUID per intended draft; reuse it only when retrying the same request.

Lists the authenticated team’s model connections with IDs, current revisions, endpoints, and verified text, structured-output, streaming, and tool capabilities. Results contain selection metadata and pagination, excluding provider credentials and raw probe output.

Creates or revises a non-executing draft for a multi-question plan inside an existing Study. Applicability: this is the setup operation for every questionnaire, survey, battery, section, cohesive question set, or request containing two or more known questions—even when the user did not say “study.” Include every question known now in this ONE draft, group related questions into cohesive named modules (question blocks or batteries), preserve their logical order, and execute them later as one confirmed run inside the Study. Never split a known set across one-question drafts or sequential runs. A one-question draft is valid only for genuinely standalone research; an adaptive follow-up whose wording depends on unavailable results can be planned later. The draft records intent, respondent-visible stimulus and questions, response formats, locale, method, outputs, confirmation questions, execution source policy, and revision metadata. Its source policy is part of the exact revision the user reviews; omitted means the controlled request_only default, while auto is an explicit experiment opt-in. Saving keeps bounded history and can discard older saved drafts or revisions. It does not start research. When the user supplies a fixed or pre-registered instrument whose wording, order, and response formats must not change, pass it as questions (one entry per item with its exact response contract) instead of request; the planner is then bypassed and the draft is an exact transcription. Explicit response contracts remain authoritative during execution, including with attachments: categoricalOptions retain their labels and order rather than being replaced by inferred A/B file labels. Answer carryover: every questionnaire item is an independent measurement. A confirmed run carries no answer history — a Mind answering item 12 has not seen its own answers to items 1-11, and items fan out in parallel — and nothing on this run can change that (historyMode is settable only on a single standalone question, and defaults to none there too). So cross-item order, fatigue, and carryover effects do not arise, and rotation or least-fill designs that exist only to control them buy nothing here; conversely, any item that must build on an earlier one has to restate what it needs in its own text or stimulus, because a back-reference resolves to nothing. Scoping stimuli to questions: attachments belong to THIS draft (one question block), not to the Study. With a single attachment, every question receives it unless questionAttachments maps that question to an empty list, which means no assets. With two or more attachments, every question needs its own questionAttachments entry (use [] for no assets) and every attachment must be assigned to at least one question, or the run is refused before any Mind answers; passing questionAttachments also drops any planner-suggested assignment from the questions you omit. Map pre-exposure, screening, demographic, and behavioural-frequency items to [] so the stimulus does not prime them. A monadic design can instead use one draft per cell whose attachments hold only that stimulus, so every question in the block receives it with no mapping to typo. That is stimulus scoping, not the forbidden question-by-question drip: each draft still carries a complete block. On a revision, omitting attachments keeps the previous set; pass an empty array to clear it. MCP cannot read or upload a local file:// path. Use a fetchable HTTP(S) URL, a signed URL supplied by the client for the attached file, or an existing Minds workspace upload URL/path. Study tools import external file URLs into durable Minds storage before saving or running. A workspace upload reused as a Study asset must sit under chat/<userId>/: request the signed upload with folder "chat" and pass that storage path or its /api/uploads/chat/ URL. A temp/, portfolio/, or context/ path — and any /api/uploads/file-access/ URL — is refused as not belonging to the workspace owner and is never re-fetched; pass a plain external URL instead. The Study refuses to start if Minds cannot read the asset. A saved Study template is a fixed instrument of exactly this kind: its stored questions are transcribed exactly as saved, with their response contracts, rather than re-planned. A template whose configuration.methodId is not guided-research carries a calculator-backed method that this transcription path does not apply, so the saved method does not run here. Before asking for confirmation, show the returned draftPlanId and numeric revision, exact respondent-visible stimulus and questions, proposed response formats, and all confirmation questions or advanced-method opt-ins. State that research has not started. A request to skip review is not confirmation of a draft the user has not seen. Wait for explicit confirmation of this displayed revision before execution.

Lists Minds research methods with availability, complexity, executable status, and fallback metadata. Results distinguish currently executable methods from experimental or planned methods.

Creates or checkpoints an unfinished Quick or Custom Study draft without starting research. It saves the objective, context, selected Audiences, method, questions, sources, and current planner step. Revisions replace the saved planning state and require the exact draft ID and expected revision; stale writes are rejected.

Applicability: use only when the user asks to find, inspect, or check a Study or its export. For a deletion-only request, do not call this or another lookup tool: Study deletion is unavailable in this app. Explain the limitation and direct the user to the Minds workspace at https://getminds.ai/ or authorized support without first retrieving the Study. Lists the authenticated user's Studies, newest first, one page at a time (limit, default 20, and offset; nextOffset continues). Each row has the Study's Audiences with their Mind counts, question count, sharing state, and workspace or shared links; a Study's Minds and results come from its status. searchQuery returns the best fuzzy name match instead of a page. A Study is the persistent workspace that contains its Audiences, questions, multi-question blocks, results, exports, and history.