Do I need the MCP server to use Peeps?
No. Peeps runs your suite on every change, analyzes failures, heals script issues, and re-verifies on its own, then reports to Slack or email. The MCP server is one of five surfaces — alongside Slack, email, Peeps AI chat, and the web app. It exists so you can drive Peeps from your editor with your local code in the same conversation.If Peeps self-heals, what is my agent fixing?
What healing hands back: a fix that needs your local code to get right, a failure diagnosed as an application bug rather than a script issue, or an intended behavior change where the steps and the script have to move together. See decide which kind of problem it is.My agent says the run is queued. Did anything happen?
Yes. The work is queued in Peeps’s hosted CI. Your agent polls and reports back when it settles.The analysis says it is still generating
Fresh failures take a moment to analyze. Your agent re-checks shortly on its own.It said nothing is failing. Is that an all-clear?
Read the sentence it leads with. An empty result names which of three things happened: failing tests exist but none inside the window asked about, nothing is currently failing, or the project has no visible completed runs. “Nothing is currently failing” means every test’s latest run passed. A test that failed earlier and has since passed will not appear no matter how far back you widen. See reading the results.How far back does it look when I ask what is failing recently?
Seven days, unless you name a range. If failures exist outside that window, the answer says so and carries the time of the most recent one, so widening takes one more question rather than a guessing ladder.Some text comes back wrapped in UNTRUSTED markers
That is your application’s output, test titles and steps, or Peeps’s own analysis text. Peeps marks it so agents treat it as data, never as instructions.My agent says it lacks permission for a tool it can see
Reconnect from Peeps settings, under Coding agents. Peeps cannot widen an existing authorization from its side — the client has to ask again. If the permission is there and the action is still refused, check your role: every change needs Contributor or above. See permissions and approvals.My agent describes a command differently from these docs
Start a new agent session. Your agent reads Peeps’ list of commands, and the instructions that come with them, when a session starts, and it can keep that copy until the session ends. Reconnecting inside the same session may not refresh it: in Claude Code, a reconnect kept the old copy and a new conversation picked up the current one. After a Peeps update, start a new session so your agent follows the current instructions.Why does my agent ask me more than once?
Each approval covers one decision. Running a fix, publishing it, and archiving something are separate, so approving one never approves the next. For runs and publishing, Peeps also ties your approval to exactly what you saw, and asks again if something changed in between.Can my agent delete things?
Very little. It can delete an empty folder and an environment variable, each after you approve. It archives test cases and script versions rather than deleting them, and you can restore those. No coding agent can delete a project.Which commands are there?
Ask in plain English and your agent picks them. To see them all, grouped by job, read the command reference.The agent asked me before running against production
By design. Run plans identify production targets and queue nothing by themselves. Review and explicitly approve before your agent proceeds.It says no Peeps project is linked to this repository
Ask your agent to list your projects, and tell it which one this checkout belongs to.My organization is not enabled for MCP
MCP is off by default and is switched on by Peeps staff, not by an org admin. Ask us to enable it. Asking your agent “who am I in Peeps?” shows which of your organizations have it on.What actually leaves my machine?
Search terms when your agent checks coverage, and the steps you approved plus a commit reference when it creates a test case. When you ask your agent to add or import a test script, it sends those files. Your checkout is never uploaded. See what leaves your machine. In the other direction, failure evidence — including screenshots, traces, videos, and DOM snapshots of your application — reaches whichever AI provider your client is configured with, under that provider’s own terms.I revoked a client. Is it locked out immediately?
Revoking stops it renewing its access, but it cannot invalidate a token already issued. That token stays valid until it expires, up to 15 minutes later. Revoking also applies to every installation authorized under that client id, on every machine.Still stuck? Get in touch.