peer-review

Peer Review MCP provides a small toolkit that lets AI agents request feedback or guidance from more capable models. It exposes two tools: ask-expert-for-peer-review and ask-expert-for-help. Use ask-expert-for-peer-review when you want a seasoned expert to evaluate your plan: supply your problem and your detailed plan, along with any relevant context. Use ask-expert-for-help when you’re stuck on a problem and want guidance from an expert by describing the problem and the issue you’re encountering. These tools help you get structured, expert input during planning and problem solving, enabling collaborative reasoning across models.

Prerequisites:

• Node.js and npm installed on your system
• Access to the root directory of the MCP server repository
• A suitable helper model and API key for the environment (see Supported Helper Models)

1. Install dependencies and build the project

npm ci
npm run build

2. Configure the MCP server in your MCP configuration file Add the following configuration (adjust /path/to/repo to the actual path where the repository is checked out, and supply your helper model and API key):

{
  "mcpServers": {
    "peer-review": {
      "command": "node",
      "args": [
        "/path/to/repo/dist/server.js"
      ],
      "env": {
        "HELPER_MODEL": "...",
        "HELPER_MODEL_API_KEY": "..."
      }
    }
  }
}

3. Replace the placeholder path and credentials with real values, then restart the MCP client (for example Claude Desktop) for changes to take effect.

4. Refer to the MCP documentation for Claude Desktop Users for integration details: https://modelcontextprotocol.io/quickstart/user

Notes:

• The server exposes two tools: ask-expert-for-peer-review and ask-expert-for-help. Ensure the chosen HELPER_MODEL is supported (e.g., Claude Sonnet/ Gemini variants) and that the API key is valid.
• If you modify the server path, keep the same structure in the mcp_config; ensure the dist/server.js output exists after npm run build.
• Restart the MCP client after changing environment variables or the server path to ensure the new configuration takes effect.
• Common issues include misconfigured paths, invalid API keys, or missing dependencies after build. Verify logs from the MCP client for troubleshooting.