Connect Replay QA to an MCP client

Connect Replay QA to a coding agent or another MCP client to browse projects, read bug reports, request testing, and verify fixes.

Use a plugin or connector

For these agents, follow their dedicated setup guides:

• Claude Code
• Codex
• Obvious

The Codex and Claude Code plugins include the MCP connection and QA workflow guidance. You do not need to add the server separately when using a plugin.

Connect another MCP client

You need a Replay account and a client that supports remote MCP servers over Streamable HTTP. QA runs use your Replay QA plan or credits.

Setting	Value
Name	replay-qa
Server URL	https://qa.replay.io/api/mcp
Transport	Streamable HTTP
Authentication	OAuth browser sign-in

1. Open your client's MCP settings and add a remote HTTP server using the URL above.
2. Start or connect the server. When your client requests authentication, complete the Replay sign-in flow in your browser.
3. Return to the client and ask it to list your Replay QA projects. The connection uses the projects your signed-in account can access.

For example, ask:

Show my Replay QA projects. Do not start a QA run.

The server advertises its tools through MCP tools/list. Clients that support MCP Apps can also display the interactive dashboard; other clients can use the tool responses to read project details and results.

Configuration examples

• Cursor
• VS Code
• Other clients

Add this server to .cursor/mcp.json, merging it with any existing servers:

.cursor/mcp.json
{
  "mcpServers": {
    "replay-qa": {
      "url": "https://qa.replay.io/api/mcp"
    }
  }
}

Connect it in Cursor's MCP settings and complete OAuth sign-in. See Cursor's MCP setup guide for its current configuration options.

Clients without OAuth support

If your client supports HTTP headers but not OAuth, create an lqa_ Replay QA API token and configure an Authorization: Bearer <token> header in the client's credential settings. See API keys and tokens for how to create the right token.

A Replay debugger API key is not a Replay QA API token. Keep tokens out of shared configuration files and chat messages.

Give your coding agent the QA workflow

A direct MCP connection exposes the tools. To give a coding agent the current setup and testing workflow, use this prompt:

Use Replay QA to test and fix this app. Start at https://docs.replay.io/llms.txt and follow it.

llms.txt is the entry point for agents. It links the Replay QA CLI workflow, the REST API (OpenAPI spec), the MCP server, the published agent skills, and these docs. Each of those describes itself, so the agent always reads the current version.

The continuous QA loop

What the agent does with it:

1. Create a project for the running app, with a short note on the flows that matter most. Optionally include a design document describing what the app should do.
2. Let QA run. Replay QA drives exploration and testing itself; the agent waits until it finishes.
3. Read the bugs. Every bug report includes reproduction steps, expected versus actual behavior, a screenshot chronology, and a root-cause analysis traced through the Replay recording down to the responsible code. The agent does not need to debug; it reads the root cause and writes the fix.
4. Mark each fix. Replay QA reruns the affected journey to confirm it.
5. Loop until no open bugs remain.

If the bug report is not enough, the agent can open the linked recording with Replay MCP and inspect the failure directly.

Apps only reachable from your machine

For an app on localhost or a private network, the agent connects it through the Replay QA CLI's reverse proxy. Testing a localhost app explains the proxy. For CI builds, see GitHub Actions or CircleCI.

Missing something?

If Replay QA lacks something you or your agent need, tell us on Discord.