@atrium/mcp-server

MCP (Model Context Protocol) server for Atrium -- exposes marketplace tools to any MCP-compatible AI agent.

This allows agents using Claude, Cursor, Windsurf, or any MCP-compatible client to discover, evaluate, and procure prediction market services directly through tool calls.

Install

npm install -g @atrium/mcp-server
# or run directly with npx
npx @atrium/mcp-server

Tools

| Tool | Description | |------|-------------| | discover_services | Semantic search across the Atrium registry | | get_service_details | Full listing details by service ID | | evaluate_service | Trigger a benchmark evaluation of a service | | procure_service | Record a procurement intent | | call_service | Call a service endpoint through the Atrium proxy | | get_reputation | Get reputation score for a service | | get_leaderboard | View the reputation leaderboard | | record_encounter | Track a service encounter for analytics | | whoami | Check the authenticated identity of the API key |

Configuration

Environment Variables

ATRIUM_API_URL=https://api.atrium.dev   # Atrium API base URL
ATRIUM_API_KEY=at_...                    # Your Atrium API key

Claude Desktop

Add to your Claude Desktop claude_desktop_config.json:

{
  "mcpServers": {
    "atrium": {
      "command": "npx",
      "args": ["@atrium/mcp-server"],
      "env": {
        "ATRIUM_API_URL": "https://api.atrium.dev",
        "ATRIUM_API_KEY": "at_your_key_here"
      }
    }
  }
}

Cursor

Add to your Cursor MCP settings (.cursor/mcp.json):

{
  "mcpServers": {
    "atrium": {
      "command": "npx",
      "args": ["@atrium/mcp-server"],
      "env": {
        "ATRIUM_API_URL": "https://api.atrium.dev",
        "ATRIUM_API_KEY": "at_your_key_here"
      }
    }
  }
}

Example Usage

Once configured, an AI agent can use Atrium tools naturally:

Agent: "Find me a sentiment analysis service for crypto markets"
→ Tool: discover_services(query="sentiment analysis crypto", domain="prediction-markets")
→ Returns: Top 5 services ranked by relevance and reputation

Agent: "Evaluate the top result against the Polymarket benchmark"
→ Tool: evaluate_service(serviceId="...", benchmark="polymarket-resolution-2024-q4")
→ Returns: Accuracy, Brier score, calibration metrics

Agent: "Call CoinGecko to get Bitcoin's current price"
→ Tool: call_service(serviceId="...", input={question: "BTC price"})
→ Returns: Service response with current data

License

MIT