Let your coding agent test
Set up once. Then your agent can test its own changes on your phones, read exactly what broke, fix it and test again. The first time, it asks you to sign in to PocketQA.
Claude Code
In Terminal, in your project:
claude mcp add --transport http pocketqa https://staging.pocketqa.dev/mcpThen in Claude Code, type /mcp and sign in. Working on the lab Mac itself? This one needs no sign-in and tests a dev server on that Mac:
claude mcp add pocketqa -- python3 ~/pocketqa/lab/mcp_local.pyClaude (web, desktop and mobile apps)
- Settings › Connectors › Add custom connector.
- Name it PocketQA, with the address
https://staging.pocketqa.dev/mcp. - Connect, and sign in when it asks.
Cursor
In ~/.cursor/mcp.json (or .cursor/mcp.json in your project):
{
"mcpServers": {
"pocketqa": {
"url": "https://staging.pocketqa.dev/mcp"
}
}
}Cursor asks you to sign in the first time it's used.
GitHub Copilot in VS Code
In .vscode/mcp.json in your project:
{
"servers": {
"pocketqa": {
"type": "http",
"url": "https://staging.pocketqa.dev/mcp"
}
}
}Start it from the file (VS Code shows Start above the server) and sign in. Use Copilot in agent mode.
CI, or anything else
Make a token in the dashboard (Agents › Tokens), then:
curl -X POST https://staging.pocketqa.dev/api/v1/tests -H "Authorization: Bearer pl_..." -H "Content-Type: application/json" -d '{"site":"www.example.com","build":"'$GITHUB_SHA'"}'Everything an agent can do is described for agents at /llms.txt.
What to ask it
- “Check what's wrong with the site on PocketQA” (it calls lab_problems).
- “Test this change on my phones before you finish” (test_site or test_url, then wait_for_result).
- “Why didn't the iPhone run?” (lab_doctor).