AI assistants and MCP
Connect AI assistants to your Autonomy projects, test cases, runs and issues.
Autonomy runs a hosted Model Context Protocol server, so an AI assistant in your editor can work with Autonomy. A connected assistant can find your projects and test cases, write new test cases, start runs against a preview, follow them step by step, read the evidence and file issues.
https://app.autonomyqa.com/api/mcpHow it connects
- Transport: one remote server over Streamable HTTP. There is nothing to install locally.
- Authorization: OAuth in the browser. The first time a client connects, you sign in and pick one organization. API keys don't work here.
- Access: the assistant acts as you, in that one organization. A member's assistant reaches only the projects that member can open. Reading and each kind of change are separate permissions you approve.
What an assistant can do
The server has 23 tools, one for each HTTP API operation: projects, test cases, test plans, runs (with Step Actions and evidence), environments and issues. Runs can target web, API and Android (iOS is coming soon), using a saved environment or a fresh preview URL or build.
The assistant doesn't click or tap anything itself. It asks Autonomy to run a test case; the agent reaches each outcome on a runner, and the assistant reads the result. See Available actions for every tool.
Use MCP when an assistant should check whether a flow already has coverage, run a test case after a change, or read a run that didn't pass before proposing a fix. The useful pattern is one coding agent changing the product and Autonomy checking it with evidence.
For scripts and CI jobs that send JSON requests instead, use the HTTP API v1.
Where MCP fits
- Dashboard: people write test cases, group them into test plans, and review runs, issues and evidence.
- GitHub: pull requests get run results and links to the evidence.
- MCP: AI assistants check coverage, start runs, follow them and read the evidence from their editor.
Get started
- Connect a client with the server URL above.
- Authorize an organization in the browser.
- Try the first MCP run on a small test case.
- Use Available actions as the reference for tool names, confirmations and following a run.
Didn’t find it?