LetsFindWhy

Agents & the API

An agent is an AI assistant that does work in LetsFindWhy on your behalf: Claude Code, a scheduled Claude routine, an editor assistant, or any other tool that speaks the Model Context Protocol (MCP). Give it an API key and it can set up projects, write discussion guides, invite respondents, and read transcripts and insights from a chat, using exactly the access you have.

Everything an agent can do, you can also do yourself with plain HTTP calls to the REST API. Both use the same key.

How to create an API key

  1. Open Settings and find the API keys card.
  2. Create a key and give it a name that says which agent will hold it, such as “Claude Code on my laptop”. One key per agent makes it easy to revoke one without disturbing the others.
  3. Copy the key straight away. It starts with mtg_ and is shown only once; if you lose it, delete it and create another.

A key is yours, not an organization’s. An agent holding it sees every organization you belong to, at the role you hold there, and everything it does is recorded as done by you. Treat the key like a password and revoke it the moment you stop using the agent.

How to connect Claude Code

The MCP server lives at https://letsfindwhy.com/api/mcp. Any MCP client that can send a header connects the same way; for Claude Code, run this once and paste your key in place of the placeholder:

claude mcp add --transport http letsfindwhy https://letsfindwhy.com/api/mcp --header "Authorization: Bearer mtg_..."

Claude Code then lists one tool per thing you can do in LetsFindWhy, named for what it does: create_project, create_discussion_guide_with_questions, invite_recruited_respondents_to_session, get_session_transcript, and so on. Ask for what you want in plain words and it picks the tools.

Cursor, scripts, and other clients that let you set a request header work the same way. The claude.ai and Claude Desktop connector directories cannot send a header; they need a sign-in flow that LetsFindWhy does not offer yet.

What an agent can and cannot do

An agent works as you. It prepares and reads research; it does not run interviews or manage your team.

It can:

  • Create and update organizations, projects, processes, discussion guides, and screeners.
  • Manage recruitment: open and close sign-up links, add respondents by hand, and invite them to sessions.
  • Set up sessions ahead of time: create them, change their settings, assign a guide, and edit its questions.
  • Read results: recordings, transcripts, responses, AI insights, and usage.

It cannot:

  • Touch a live interview. Starting, ending, admitting people, removing them, and summoning or dismissing the AI moderator stay with the people in the room.
  • Change who is in your organization: it cannot change roles, remove members, transfer ownership, leave, or delete the organization.
  • Delete recordings. Deleting a recording also deletes its transcript, so that stays a decision you make in the app.

The line is drawn where it is because an agent reads what respondents said, and a respondent could say something designed to steer it. Whatever it reads, it cannot lock you out of your own organization or erase your research.

Deleting a project, session, process, guide, or screener is allowed. Projects and sessions can be restored for 30 days, and your agent will ask you to confirm before any delete.

How to call the REST API directly

Every tool is a REST endpoint underneath. Send your key as Authorization: Bearer mtg_... and browse the full reference, with every endpoint and its parameters, at /api/docs once you are signed in.

Keys are limited to short bursts: 60 calls in one second over the REST API, and about 30 through MCP, where each tool call checks the key twice. No thinking agent reaches either. A call that goes over gets a 429 response saying how long to wait.