MCP

MCP (the Model Context Protocol) lets AI tools such as ChatGPT, Claude, Claude Code and Cursor use other services. User Evaluation's MCP server lets them read your studies, transcripts, study reports and reports, search what participants said, ask Eva, and draft a new study.

• Server address: https://api.userevaluation.com/mcp
• Transport: Streamable HTTP. There is nothing to install.
• Sign-in: OAuth (you sign in to User Evaluation and approve the client), or an API key sent as Authorization: Bearer ue_live_….

Connect ChatGPT or Claude

ChatGPT and Claude (on the web and in Claude Desktop) connect as a custom connector. You don't need an API key.

1. ChatGPT: go to Settings → Apps & Connectors, turn on developer mode if asked, and choose Create. Claude: go to Settings → Connectors and choose Add custom connector.
2. Name it User Evaluation and enter https://api.userevaluation.com/mcp as the URL. In ChatGPT, choose OAuth for authentication.
3. A User Evaluation page opens. Sign in if you need to.
4. Choose the workspace to connect (if you're in more than one) and what the assistant may do:
   • Read your research and ask Eva (always on).
   • Draft studies: create study drafts, and let Eva make changes once you approve them in the app. Observers can't give this.
5. Choose Allow. You're sent back to ChatGPT or Claude, and the connector is ready.

The connection shows in Settings → API & MCP → Connected, with who connected it, what it may do and how many calls it made.

Connect Claude Code

claude mcp add --transport http userevaluation https://api.userevaluation.com/mcp

Then run /mcp in Claude Code and choose userevaluation. Your browser opens the same sign-in and approval page.

Connect with an API key

Cursor, VS Code and other clients that can send a header can use an API key instead.

1. Go to Settings → API & MCP and create an API key. Tick Read only if the assistant only needs to read.
2. Copy the key. It is shown only once.
3. Add the server to your client. Put your key in place of ue_live_….

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "userevaluation": {
      "url": "https://api.userevaluation.com/mcp",
      "headers": { "Authorization": "Bearer ue_live_…" }
    }
  }
}

VS Code (.vscode/mcp.json)

{
  "servers": {
    "userevaluation": {
      "type": "http",
      "url": "https://api.userevaluation.com/mcp",
      "headers": { "Authorization": "Bearer ue_live_…" }
    }
  }
}

For client developers: OAuth details

The server follows the MCP authorization spec:

• A request without a valid token gets 401 with WWW-Authenticate: Bearer resource_metadata="https://api.userevaluation.com/.well-known/oauth-protected-resource/mcp".
• Protected Resource Metadata (RFC 9728) is at /.well-known/oauth-protected-resource and /.well-known/oauth-protected-resource/mcp. Authorization Server Metadata (RFC 8414) is at /.well-known/oauth-authorization-server (also served at /.well-known/openid-configuration).
• Clients register with Dynamic Client Registration (RFC 7591) at POST /register. Clients are public (token_endpoint_auth_method: none). Redirect URIs must be https, http on localhost/127.0.0.1, or an app's own scheme, and must match exactly at /authorize.
• GET /authorize takes the authorization code flow with PKCE S256 (required) and the resource indicator (RFC 8707). The redirect back carries code, state and iss.
• POST /token supports authorization_code and refresh_token. Access tokens last one hour. Refresh tokens last 30 days and work once: each refresh returns a new pair. Using an old refresh token again revokes the connection.
• POST /revoke (RFC 7009) revokes the connection a token belongs to.
• Scopes: read (read and ask Eva) and write (draft studies, Eva may change things with approval in the app). Tokens only work on the MCP server, not on the REST API.

Tools

Tool	What it does
list_studies	Lists studies, newest first. Filter by status or type, or search with q.
get_study	One study: question, guide, screener, audience and status. Takes an id or a slug.
get_study_status	A study's status and progress: invited, booked, completed, quality, next session.
list_sessions	A study's sessions, without transcripts.
get_session	One session with its full transcript, chapters and signals.
get_readout	The study report for a study: themes, claims with their clips, next steps.
list_reports / get_report	Reports in the workspace, and one report with its document.
list_swarm_runs	Swarm runs, newest first.
list_findings / get_finding	The findings of a Swarm run, and one finding with its evidence.
search	Finds studies, reports, chats, Library files, projects, Swarm runs and people by name.
search_evidence	Searches what participants said: transcripts, clips, Library transcripts and study report claims.
list_library_files / get_library_file	Library files, and one file with its transcript.
ask_eva	Asks Eva. The answer cites its evidence and is saved as a chat in the app.
create_draft_study	Drafts a study from a research question. Nothing is published or charged; you review and launch it in the app.

What a connection or key can do

The server uses the same rules as the API. A connection or key acts as the person who made it, with their workspace role, up to Member rights, and only sees that workspace.

• Read-only keys and connections without Draft studies can use every tool that reads, and can ask Eva. They can't draft a study. Eva can't change anything for them.
• Observers can read and ask Eva, but can't draft a study.
• Revoking a key in Settings stops it at once, for the API and for MCP.
• Revoking a connection in Settings → API & MCP → Connected stops its access and refresh tokens at once. The person who connected it or a workspace admin can revoke it.

Before Eva spends money or contacts people, she asks for approval in the app.

The MCP server, like the API, comes with the Pro, Team and Scale plans. On Free you can't create a key or approve a connection, and keys and connections made on a paid plan stop working (402) until you upgrade again.

Calls to the MCP server count towards the workspace's API calls. They show as MCP in the Usage chart in Settings. Each key can send 600 requests a minute, and the workspace's monthly calls are the same as for the API.