Skip to main content

​
Where the MCP server fits

Peeps runs on its own: it executes your suite on every change, analyzes and heals failures, and reports to Slack or email. None of what follows is required for that to happen. The Peeps MCP server is the surface for when you want to drive Peeps from the terminal you already work in. Your coding agent reaches the same failure evidence, the same runs, and the same test cases — alongside your local checkout, so your own code is the context for every diagnosis and every proposed fix.

​
What you can ask for

The guides are in the order most people need them.

Troubleshoot and fix a failure

Open a failure in full — cause, error, script, screenshot, history — then fix it and publish the verified fix.

Run tests

Run a test in Peeps’s hosted CI against the environment you name, after you approve the plan.

Find tests and check coverage

Search your test cases, read one in full, and check what the code in front of you already has covered.

Create and import tests

Create cases for real gaps and generate their scripts, or bring in your existing Playwright tests.

Manage tests

Edit details and steps, set what runs first, archive and restore, and keep folders tidy.

Environments and projects

Set up and rename environments, edit their variables, and create or rename projects.
You never name a tool. You say “why is the checkout test failing?” and your agent works out which Peeps project you mean, finds the failing run, and pulls the whole evidence bundle in one call. To see every command behind that, read the command reference. It lists them by job.

​
What you need

  • A Peeps organization with MCP enabled. It is off by default and switched on by Peeps staff. Asking your agent “who am I in Peeps?” reports which of your organizations have it on.
  • Claude Code, Cursor, or Codex.
There is no package to install and no API key to manage. The connection is OAuth, approved in your browser, and it is per machine and per client — not per repository. Connect once and every checkout on that machine can use it.
Troubleshooting and coverage checks never upload your source code. A coverage check sends search terms only. Creating a test case sends the steps you approved plus a commit reference. The exception is a test file you explicitly ask your agent to add or import: that file is what you are sending. See what leaves your machine.

Connect your agent

Setup takes about two minutes, once per machine.