MCP Integration

Connect UserTold to an MCP-compatible agent so it can set up research, read interview evidence, and prepare evidence-backed work.

Fast path: let your agent set it up

Help me connect UserTold to this agent with https://mcp.usertold.ai/mcp and complete browser authorization.

Inspect my workspace before creating anything. If it is empty, ask what product I am researching and what I want to learn. Then create my first project, a participant intake, and a draft interview study linked to that intake. Show me the study and intake for review before activating either one. After I approve them, activate the study and linked intake, then hand me off to the dashboard to copy the final widget embed and tell me the next step for testing it. If setup already exists, summarize it and help me continue from the current state.

Connect

Add this remote HTTP server to your MCP client:

https://mcp.usertold.ai/mcp

Your client opens a browser login using OAuth 2.1 with PKCE. You do not need to copy an API token.

Discover the current surface

After authentication, let MCP describe what is available to the agent:

  • tools/list returns current actions and input schemas.
  • resources/list and resources/templates/list return readable product context.
  • prompts/list returns reusable research workflows.

Some tools load on demand. If a tool is not in tools/list, read usertold://mcp/deferred-tools, then read the schema URI it returns before calling the tool.

The initialize response and usertold://me identify the authenticated workspace. Use project IDs returned by UserTold for follow-on calls instead of reconstructing references.

Product boundaries

  • Activation requires your explicit approval.
  • Read source evidence and interview context before turning a review packet into delivery work.
  • Copy and visually verify the final public embed in the dashboard.
  • work.push sends verified work to GitHub or Linear; it does not decide whether the work is ready.

If setup fails

  • Login does not complete: reconnect the server and finish the browser authorization flow.
  • A tool is missing: read usertold://mcp/deferred-tools and the returned tool schema.
  • A project cannot be found: read usertold://me and use a project ID returned by UserTold.
  • The embed is unavailable: confirm the study and linked intake are active, then open the project in the dashboard.

Continue