Skip to main content
The QA.tech MCP server exposes your test cases, test plans, runs, issues, chat conversations, automations, configs, knowledge base, and agent rules to any AI client that speaks the Model Context Protocol. Once connected, your assistant can answer questions like “what runs failed today?” or “rerun the failed cases from run UkxK” without you leaving the chat.

Set up with your agent

Paste this prompt into Claude Code, Cursor, or Codex. Your agent signs you up (or signs you in), connects the MCP server, and runs your first tests. It stops to guide you through the steps that happen in your browser.
Prompt
The agent follows the Agent Quickstart. To set it up by hand instead, read on.

Accessing the MCP

The MCP is available at:
Authentication is handled via OAuth.
  • On first connect you sign in to QA.tech and pick an organization (and optionally specific projects).
  • You can only access data you have permission to view in that organization, limited to the projects you approved.
API keys still work for CI and clients that don’t speak OAuth yet. Open the dashboard at Organization Settings → MCP Server for copy-paste client snippets.

Pick Your Client

Most modern clients have a one-click or one-line install. Clients that support MCP OAuth (Claude Code, Cursor, Codex, ChatGPT, Linear Agent) only need the URL above. API keys still work for CI and clients that don’t do OAuth yet.
Run this in any project where you use Claude Code. Requires the Claude Code CLI.
Then run /mcp in Claude Code (or claude mcp login qatech) and complete the browser sign-in. Claude Code reloads MCP servers automatically.Prefer a static key instead? Add --header 'Authorization: Bearer <API_KEY>' from Organization Settings → API Keys.

Or Use the QA.tech CLI

The QA.tech CLI can write the config for any supported client in one command. Useful if you switch machines often or want the same setup scripted.
Don’t have the QA.tech CLI? Install it with npm install -g @qadottech/cli.
For Cursor and Claude Desktop, the default is a URL-only OAuth entry (no API key):
Clients that still need a Bearer token (Continue, Goose) require an API key first:
Pass --api-key to force a static Bearer entry for Cursor/Claude Desktop. Add --print to dump the snippet without writing it. See qatech mcp configure --help for all flags.

What Your Assistant Can Do

The server exposes 59 tools: 34 that need read scope and 25 that need write scope. Tools your credential lacks the scope for are hidden from tools/list, so a read-only API key never sees the write tools. OAuth grants include both scopes. MCP clients that group tools by risk or ask before acting (Claude Desktop Tool Permissions, ChatGPT) read the protocol annotations, not the scopes. readOnlyHint follows the scope. destructiveHint is set only on tools that delete, overwrite, or cancel something (the manage_* entity tools, update_*, cancel_run). openWorldHint is set on tools that drive a browser against your site: starting or rerunning runs, chat, triggering automations, and create_test_case, which starts a burn-in run. Each entity has one write tool, manage_<entity>, that takes an operation of create, update, or delete. Its reads stay in separate read-only tools (list_*, get_*), so clients can auto-approve them. Field descriptions say which operations use or require each field. The earlier per-operation tools (create_rule, update_rule, delete_rule, and the same for knowledge items, automations, configs, labels, test plans, and environments) are hidden from tools/list but still callable, so existing agents keep working. Most tools are project-scoped and take a projectShortId. Call list_projects first to find it, or use a project-scoped API key, which falls back to its own project.

Runs and results

Test cases

Issues

Todos

Read-only API keys do not see manage_todos. They can list and fetch todos with GET /v1/todos.

Projects, applications, and environments

Mobile builds

These tools back mobile app testing. Upload the file to the presigned URL yourself with an HTTP PUT between the first two calls.

Knowledge base

Rules

Test plans

Labels

Scenarios

Configs

Features

Chat

Automations

See the Automations API for the trigger and action options.

Remote tunnels

Documentation

Feedback

Example prompts

“Which of my test cases failed in the last run?”
“Rerun run UkxK, but only the failed cases.”
“Show me the agent’s trace for the failing checkout test and tell me where it went wrong.”
“List the critical issues QA.tech found this week, then help me fix the first one.”
“Create a test case titled ‘Checkout with expired card’ for the frontend app.”
“Add our new API error codes to the knowledge base so the agents stop reporting them as bugs.”
“Which features cover inviting a teammate?”
“How do I connect the QA.tech MCP server in Cursor?”

Authentication

Two options: Claude Code, Cursor, Codex, Linear Agent, and other MCP clients that support OAuth discover QA.tech’s authorization server automatically. On first connect you sign in with your QA.tech account, pick an organization, and approve access. Tokens refresh in the background. OAuth grants include both read and write, so every tool is available for the organization and projects you approved.

API key

Same Bearer tokens as the REST API. Useful for CI and clients that don’t speak OAuth yet.
  • Where to find your key: Organization Settings → API Keys in the dashboard.
  • Scopes: read and write are independent. A key with read sees the 29 read tools, a key with write sees the 23 write tools, and a key with both sees all 52.
  • Project binding: Keys may be org-scoped or project-scoped. Org-scoped keys need a projectShortId on project-scoped tool calls; project-scoped keys fall back to their own project.
Treat API keys as confidential credentials. Anyone with the token can read your test data and, with a write key, create and rerun tests. Rotate keys from the dashboard if one leaks.

Endpoint Details

Batch JSON-RPC requests and GET upgrades are not supported. Each call is a single POST with a JSON body.

Smoke test with curl

A successful response lists the tools your key has access to.

Troubleshooting

For anything else, contact support.