Minds MCP Server for ChatGPT, Claude, and Cursor

The server currently advertises 50 tools and keeps 77 canonical tools callable, read from the live server. Treat the connected server’s tools/list response as authoritative.

Minds provides a Model Context Protocol (MCP) server that lets AI assistants create reusable Audiences, run cohesive multi-question Studies, and interact with individual Minds. This enables integration with ChatGPT, Claude, Codex, Gemini CLI, Cursor, VS Code, and other MCP-compatible clients.

What is MCP?

The Model Context Protocol (MCP) is an open standard that enables AI assistants to securely connect to external tools and data sources. With our MCP integration, you can:

Create and manage AI Minds (experts, personas, digital twins)
Build Studies with reusable Audiences of domain experts or consumers
Run survey-style questions across multiple perspectives
Get statistical analytics on Study responses
Export branded PDF reports, markdown summaries, or raw data
Organize Minds into reusable Audiences
Create grounded audiences from briefs or respondent datasets
Plan, explicitly confirm, run, and resume durable research Studies
Manage knowledge, Formations, lifecycle actions, and stateful chats

MCP Server URL

https://getminds.ai/mcp

For setup instructions by client, see the Minds MCP setup guide. For every canonical tool and the live advertised surface, see the Minds MCP tools reference. Agent authors should also read the MCP operating guide for agents.

Authentication

The MCP server supports two authentication methods:

OAuth 2.1 (recommended) — Clients like ChatGPT, Claude, Codex and Gemini CLI handle this automatically. You'll be prompted to log in to your Minds account when connecting. The setup guide lists the permissions a client requests.
API Key — For programmatic access, generate a key at Settings → API Keys (starts with minds_). Pass as Authorization: Bearer minds_your_key.

Compatible Clients

Client	Widget	Auth	Setup
ChatGPT	Client-dependent	OAuth	Web: Plugins → Add → Create MCP App, or Minds from the plugin directory; desktop app: Plugins → Add → Add MCP server (Streamable HTTP, then Save → Restart → Authenticate)
Claude (claude.ai and Claude Desktop)	Client-dependent	OAuth	Customize → Connectors → + → Add custom connector
Claude Code	No	OAuth or API key	`claude mcp add --transport http minds https://getminds.ai/mcp`, then `/mcp` → Authenticate
Codex	No	OAuth or API key	`codex mcp add minds --url https://getminds.ai/mcp`, then `codex mcp login minds`
Gemini CLI	No	OAuth	`httpUrl` entry in `~/.gemini/settings.json`
Cursor	Client-dependent	OAuth	`url` entry in `~/.cursor/mcp.json`
VS Code (GitHub Copilot)	Client-dependent	OAuth	MCP: Add Server → HTTP, then the server URL
Langdock	No	OAuth	Integrations → Add MCP
Windsurf	No	OAuth or API key	Cascade … → Open MCP config file, then a `serverUrl` entry

Every client sees the same advertised tools. Step-by-step instructions for each client are in the setup guide; the ChatGPT guide walks through every ChatGPT screen.

All clients receive clickable links to open results in the Minds webapp, regardless of widget support.

Advertised versus canonical tools

tools/list returns 50 curated tools designed for ordinary interactive research. The server keeps 77 canonical tools callable in total, including lifecycle tools it does not advertise. Explicit integrations can call those tools by canonical name when the client supports configured or direct tool calls. This split keeps the default model surface focused while preserving the full supported capability contract.

Tool aliases may remain callable for compatibility, but agents should always use canonical names from the tools reference.

Area	Tool	Description	Behavior
Audiences	`list_audiences`	List Audiences. For a listing or search reply, include both the returned top-level workspaceUrl and each listed Audience's workspaceUrl verbatim in inline code: scheme, query and trailing slash. Item links do not replace the list-level link. Lists owned Audiences newest first with counts and sharing. workspaceUrl opens the owner's workspace; sharedAudienceUrl is public. addOnRequired omits links/commercial details; use an available Audience. First-page openDrafts: rework matching drafts or recalibrate matching Audiences before a new brief.	Read-only
	`import_audience_sources`	Import Audience Research Sources. 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.	Writes
	`create_audience_from_brief`	Create a Grounded Audience from a Brief. Builds grounded Audience/Mind profiles with exact segment allocation. Returns an asynchronous draft/operation for review; create only after the user accepts its roster/quota snapshot and checksum. Rework the same draft.id/sha256 in place, overwriting its saved request, roster and grounding; never start a new brief for that edit. Supports reviewed datasets and public research/files. Existing Audiences use recalibration. Private unless link sharing is enabled.	Destructive
	`get_audience_creation_progress`	Audience creation progress. Read-only; never starts or retries work. operationId: creation and member training. audienceId: training, background research, validation.ready (validation's 10-ready-Mind rule), validation.reason MINDS_TRAINING (poll later) or AUDIENCE_TOO_SMALL (add Minds). recalibrationId: its status, retained result or terminal error. Final score needs all members trained and research done.	Read-only
	`render_audience_review`	Show Audience Creation. Shows source research, distributions, proposed Minds, the draft and training for one or more Audience operations, display only: the user confirms or asks for changes in chat. Multiple operations appear in independent tabs within one widget.	Read-only
	`get_audience_limits`	Get Audience Limits. Returns the authenticated account's current plan (`planName`, as the app names it), its Audience size allowance (per-Audience cap including any team or contract allowance, `customSizeMax`, per-mode ceilings) and the synthetic responses left, read live; it supersedes limits stated elsewhere. Sample sizes the plan allows: up to `customSizeMax` Minds per Audience, or several Audiences in one Study for a larger sample, each costing one response per Mind per question.	Read-only
	`ask_audience`	Ask One Standalone Audience Question. Accepts one standalone question request for an Audience, restricted to supplied segments, and creates a private Study. Returns IDs/links. For preparing, lead with userFacingStatus and call it a request throughout.	Destructive
	`export_audience`	Export Audience Brief. Exports an Audience brief or kind validation_report (scores, KPIs, answer shares, provenance) as Markdown/PDF/DOCX/PPTX. Brief PDF/PPTX can fetch public HTTPS member portraits. Starts artifact work; force can replace it. Slow exports may irreversibly email the requesting account.	Destructive
	`duplicate_audience`	Duplicate Audience. Adds an independent copy of an Audience, its Minds and retained knowledge. Unready copies can queue retraining with public web research when retained evidence cannot seed them. Ready copies avoid retraining.	Writes
	`recalibrate_audience`	Change an Existing Audience. Changes an existing Audience in place: it keeps its id, its Minds keep theirs (except copies below), and no Mind is deleted. Covers composition and trait edits, a population pivot, adding members, an exact chart selection (`distributions`), new sources and going deeper (`deepen`), and re-runs its research grounding. Owner, or verified Minds staff with existing direct Audience admin collaboration or an active owner-issued support grant. Staff support is limited to composition amendments: no growth, new sources, Go deeper or shared-Mind waiver. Asynchronous: the result is an accepted operation with its id, not a completed change. Queues behind another edit of the Audience and waits for a Study run on its members. Members shared with other Audiences are replaced here by copies with new ids. Reusing requestKey with identical input retries safely.	Destructive
	`manage_audience_support`	Manage Audience Support Access. Manage owner-issued, expiring staff access to one Audience’s composition and receipts. Staff cannot self-grant. Revocation stops new work at checkpoints; completed changes remain. No membership, growth, deletion or billing changes.	Destructive
	`attach_study_audiences`	Attach Audiences to Study. Attaches existing accessible Audiences without removing others, planning or running research. Optional segmentSelections retain exact matches. Returns attached/alreadyAttached IDs; retries are safe. Link sharing makes these Audiences and Minds public. Validates access and plan eligibility.	Writes
Studies	`list_studies`	List Studies. List or search saved Studies, or read an authorized team catalog, result feed or tracker snapshot with source identities, full-filter totals and page continuation.	Read-only
	`create_study`	Create a Separate Study Workspace. Creates a separate, new Study workspace when the user requests a new workspace. It can attach existing Audiences or create Audiences inline. It does not ask questions or run research. It does not act in an existing Study or save a draft there.	Writes
	`ask_study`	Ask One Standalone Question in a Study. Requests exactly one respondent-visible question: one standalone question, or one adaptive follow-up whose wording could not be known before earlier results. Out of scope: questionnaires, batteries, sections or two or more known questions; use one confirmed block, submitted once rather than question by question. Durably accepts before execution. planning_required requires confirmation. Question text can reach the Minds; exclude planner/client notes. Pasted material: verbatim in evidence.textAttachments. Accepts files shared in the conversation (uploaded or generated images, PDFs). Returns IDs/links. Fixed time-bucket retries can submit again; check status before repeating uncertain writes.	Destructive
	`get_study_status`	Get Study Status. Read Study progress, saved replies, uncalibrated forecasts and exports without generating replies. Scope by runId or questionId (including a preparation clientRequestId). A preparing request has userFacingStatus and no submitted question.	Destructive
	`request_study_answers`	Request Study Replies. Request actual replies (default 8; all=true for all). May spend response allowance.	Destructive
	`render_study_results`	Show Study Results. Completed Study results or labeled experimental, uncalibrated forecasts. Exactly one runId/questionId. Replies load on request. May recover and replace derived results from accepted answers without rerunning Minds.	Destructive
	`export_study`	Export Study Results. Queues Study results export: executive/full reports in PDF/DOCX/PPTX/Markdown; raw CSV/XLS/SPSS SAV. Returns job/artifact status. Full reports can fetch public HTTPS portraits. force can replace artifacts. Completed exports can irreversibly email the requesting account.	Destructive
	`duplicate_study`	Duplicate Study. Permanently copy a Study with its questions and results over the same Audiences.	Writes
	`export_heatmap`	Export Website Heatmap. Exports a completed website heatmap by Study/message ID as a ZIP of stored screenshots and overlays. Does not recapture the website. force starts fresh archive work and replaces its saved artifact status/reference.	Destructive
	`get_study_heatmap`	Read Study Asset Heatmap. Read saved attention by assetKey and, for monadic Studies, variantId. No new analysis, website visits, or response charges.	Read-only
	`start_study_heatmap`	Start Study Asset Heatmap. Start or retry analysis of an assigned original file, image, video or public website, including a monadic variantId. Mosaic prepares original files internally and retains saved material findings and visual groups. Eligible Mosaic Reactions start automatically with the Study and use no additional response credits. Other paid or legacy analysis requires approval, a paid plan and one response per Mind. Preserves question answers.	Destructive
	`create_study_plan`	Draft Research in an Existing Study. Saves a non-executing research draft in an existing Study, including a multi-question concept test or survey. It creates no Study workspace and starts no research. study.name directly resolves the existing Study and its Audiences by name. study.id selects an ID; omission uses the active Study.	Destructive
	`revise_study_plan`	Revise Study Plan. Revises the saved draft at current draft.id and draft.revision without starting research. Nest changes in draft.refinement, answers in draft.answers, formats in draft.questionResponses, or exact edits in draft.edits. Replacement questions are top-level questions. Stale revisions fail; answers alone preserve fixed wording. Review the new revision. History is bounded.	Destructive
	`get_study_plan`	Read Study Plan. Reads the latest saved revision and full review of one draft. Read-only; starts no research.	Read-only
	`render_study_plan`	Show Study Plan. Shows a saved Study draft at its current revision for interactive review and confirmation. Research has not started.	Read-only
	`cancel_study_question`	Stop Standalone Study Question. Stops one standalone Study question by runId, or studyId and questionId: queued or preparing stops at once, being answered stops at its next step. Answers given are kept. A delivered forecast keeps its completed run history while remaining replies and Reactions can still be stopped. An answered question with no pending work, or a failed question with no work stopped, errors unchanged. Not for multi-question runs.	Destructive
	`read_study_summary`	Read Study Summary. Reads saved whole-Study Markdown, evidence, generation state and your refresh receipt. Pending retains prior content. May recover/replace stale or missing summaries; never reruns respondents.	Destructive
	`refresh_study_summary`	Refresh Study Summary. Queues a whole-Study summary refresh after explicit approval. Replaces it without restore history; preserves questions/answers. force regenerates unchanged coverage. Pending returns a receipt and prior content.	Destructive
Research planning	`cancel_study_run`	Cancel Study Run. Stop pending Study work; keep saved replies. Returns run/collection status.	Destructive
	`start_study_run`	Start Confirmed Study Run. Runs one reviewed draft revision after explicit confirmation in a later user turn; skipping review is not confirmation. One execution submits the whole draft as a single durable run. It is refused as a whole before using Minds or quota if the revision, capabilities, method or source-readability checks fail. Advanced methods need explicit opt-in. Each Mind answers in order seeing its own earlier answers; askIf gates items. Contradictions are flagged, never rewritten.	Destructive
	`resume_study_run`	Resume Study Run. Resumes an existing plan-limited Study run in place after explicit user approval and the required paid-plan entitlement. Keeps the originally confirmed questions and response contracts and does not ask questions already answered. Does not start a new draft or change the run design.	Destructive
	`list_research_methods`	List Research Methods. Lists the research-method catalog without changing or running methods. For availability, report every label and the exact count in each availabilitySummary group. Use methods for complexity and fallbacks; advanced methods can require opt-in.	Read-only
	`plan_study_questions`	Plan a Multi-Question Block in a Study. Deprecated alias of create_study_plan and revise_study_plan; identical behaviour.	Destructive
	`run_study_questions`	Run a Confirmed Multi-Question Block. Deprecated alias of start_study_run; identical behaviour.	Destructive
Templates & drafts	`list_study_drafts`	List Study Drafts. Lists a page of unfinished Study draft summaries, newest first, or returns the complete saved planning state for one exact draft ID. Draft records are distinct from running or completed studies.	Read-only
	`list_study_templates`	List Study Templates. Lists owned/team-shared template summaries, most used first; templateId returns full configuration, revision, questions, response settings and attachments. configuration.questions is the fixed instrument; configuration.methodId the research method it runs with. Listing provisions starters or overwrites eligible untouched older starters. Exact-template reads skip provisioning.	Destructive
	`save_study_draft`	Save Study Draft. 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 change only the fields given and require the exact draft ID and expected revision; stale writes are rejected. For a retryable creation, choose idempotencyKey before the first save and reuse it after an uncertain result.	Destructive
	`save_study_template`	Save Study Template. Creates a private Custom template with its questions, response settings, attachments and method. template.requestId enables safe retries. Does not change existing templates, create a draft or start research.	Writes
	`update_study_template`	Update Study Template. Updates one owned Custom template at update.expectedRevision. Replaces supplied fields without restore history and may share it with the team. Requires explicit approval; stale revisions fail. Does not start research.	Destructive
	`use_study_template`	Create Draft From Study Template. Creates an independent editable Study draft from a readable template at use.expectedRevision. Reuse use.requestId for retries of the same intended draft. Fresh context and Audiences are added in the web app; the template is unchanged and research does not start.	Writes
	`delete_study_template`	Delete Study Template. Only for saved Study templates; it cannot delete Studies, Audiences, Minds or answers. Permanently deletes one template the authenticated user owns; Studies and drafts already created from it are unaffected.	Destructive
Minds	`export_mind`	Export Mind Persona Profile. Exports an existing Mind by ID/name: inline Markdown by default, or PDF/DOCX/PPTX. PDF/PPTX can fetch public HTTPS portraits. Starts artifact work; force can replace an artifact. Slow exports may irreversibly email the requesting account when ready.	Destructive
Minds	`get_shared_mind_knowledge`	Read Shared Mind Sources. Returns stored public Mind sources/assessments by share ID, including shares outside the private workspace. Does not fetch source websites or modify knowledge.	Read-only
Other	`list_formations`	List Audience Segmentations. Lists Audience segmentations for filters; may save defaults and recover builds, replacing subgroup memberships. Ready counts are final. Paged (limit 1–100, default 100); follow pagination.nextOffset.	Destructive
	`update_item`	Update Study, Audience or Mind. Renames a Study or Audience or moves it into a folder. Renames a Mind, or records whether its facts come from real research data (mind.factsSource).	Destructive
	`set_sharing`	Change Study or Audience Sharing. Turns a Study's or Audience's public link or team sharing on or off.	Destructive
	`enterprise_use_cases`	Enterprise use cases. Read paged team business use cases and role-scoped counts of attributed drafts, Studies and completed runs.	Read-only
	`save_enterprise_use_case`	Save enterprise use case. Save a revision-checked team business use case with objectives, template revisions, planned targets and lifecycle state.	Destructive

Interactive Preview

Minds currently registers three MCP views: Study results (for a new question and for an existing Study), research-plan review, and Audience creation. They reuse Minds UI components. The client determines the surrounding frame and available interactions.

The current plan view displays the saved draft for review. Request changes in the conversation, then explicitly confirm the latest revision before the assistant runs it. Viewing a plan does not start a Study.

When a widget detects a mobile host, it displays a link to continue in Minds; interactive controls and widget polling do not start. This changes the widget presentation, not the MCP server's tool availability. Client restrictions still apply.

The examples below use illustrative data. Each tab is one view, with the tools that open it; loading and in-progress examples are fixed snapshots, not live research.

3 widgets · 13 preview states

Inspect every MCP widget

Shown by: render_study_results

The first render, before the host delivers the question's data.

Illustrative data · appearance varies by client

n8n workflows

Building an n8n automation instead of an MCP client? The Minds n8n integration connects directly to the Study API and can serve as an AI Agent tool. The npm package supports self-hosted n8n while n8n verification remains under review.