RedCrown · MCP Server

The RedCrown prove-loop, as tools for your agent.

Check a changed AI feature against a reference run and your written requirements, and keep the evidence for each case. This endpoint gives Claude and other agents access to your RedCrown records over the Model Context Protocol.

This is a machine endpoint, not a web app. Agents connect to it over MCP; the protocol lives at POST https://mcp.redcrown.ai/mcp and requires an OAuth login with your RedCrown account. The server is a stateless shim: it validates your token and forwards it to the RedCrown API, so it holds no secrets and all data stays scoped to your account. Looking for the product? Visit redcrown.ai.

The two tools most callers need

`prove_task`

Give a plain-language task and a few examples. RedCrown runs several models, ranks them against your quality bar, and returns the ranking with the report's caveats. When the scoring method is not clear, it runs nothing and asks you to confirm a method with quality_metric. Pass publish: true to also create a share link. Leave the expected output blank to rank against the model you use now.

`try_sample`

Returns a stored example report, with its ranking and caveats. It needs no input, no keys and no setup, and it starts no new run.

Sixteen more advanced tools drive the full loop (import results, scaffold and run experiments, live proxy capture, and the reviewer decision report).

Connect your agent

Claude (Desktop / claude.ai): open connector settings, add a custom connector, enter https://mcp.redcrown.ai, and approve the OAuth login.

Claude Code:

claude mcp add --transport http redcrown https://mcp.redcrown.ai

Cursor, Codex, and other mcp.json clients:

{
  "mcpServers": {
    "redcrown": { "type": "http", "url": "https://mcp.redcrown.ai" }
  }
}

Once connected, try: "Use RedCrown to rank models for classifying these support tickets, and publish the result." The agent calls prove_task with publish: true and gives you the share link.

redcrown.ai Source on GitHub OAuth metadata