AI Assistants and MCP

Autonomy publishes a hosted Model Context Protocol server so AI assistants can operate the QA control plane from an editor without becoming the browser or device driver. A connected client can find existing Test Cases, organize them into Test Plans, trigger Runs against preview targets, inspect progress, and retrieve External Evidence.

https://app.autonomyqa.com/api/mcp

How it connects

• Transport — one remote server over Streamable HTTP. There is no local install or stdio process to manage.
• Authorization — OAuth 2.0 in the browser. The first time a client connects, Autonomy opens a sign-in window where you pick the organization that owns the Test Cases and Runs.
• Scope — read access and each resource's write access are separate OAuth scopes. Every request is limited to the selected organization.

What an assistant can do

The server publishes 15 tools for Test Cases, Test Plans, Runs, Environments, and External Evidence. Notes remain HTTP-only. A client can trigger web, API, iOS, and Android Runs using a saved Environment or explicit targets from CI and preview systems. It cannot issue raw browser clicks, drive a native device directly, or perform arbitrary network execution. See Available Actions for the complete catalog.

Use MCP when an assistant should check whether a flow already has coverage, run one Test Case or a Test Plan after a change, or inspect a failed Run before proposing a fix. The useful pattern is one coding agent changing the product and an Autonomy-backed QA loop validating it with evidence.

For scripts and CI jobs that need direct JSON requests instead of an MCP client, use the HTTP API v1.

Where MCP fits

• Dashboard — humans author Test Cases, group them into Test Plans, and review Runs, Environments, and evidence.
• GitHub — pull requests receive Run status and evidence links.
• MCP — AI assistants inspect coverage, trigger QA, wait for Runs, and retrieve External Evidence from their editor or agent runtime.

Get started

1. Connect a client with the server URL above.
2. Authorize an organization in the browser.
3. Run the First MCP Run workflow against a preview URL, API base URL, or saved Environment.
4. Use Available Actions as the reference for tool names, confirmation rules, Run waiting, and trigger target fields.