Enable AI assistants like Claude Code, Claude Desktop, and Cursor to reflect on, critique, and continuously improve their own performance using Mandoline's evaluation framework via the Model Context Protocol.
Most users should start here. Use Mandoline's hosted MCP server to integrate evaluation tools into your AI assistant.
For each integration below, replace sk_**** with your actual API key from mandoline.ai/account.
Use the CLI to add the Mandoline MCP server to Claude Code:
You can use --scope user (across projects) or --scope project (current project only).
Note: Restart any active Claude Code sessions after configuration changes.
Verify: Run /mcp in Claude Code to see Mandoline listed as an active server.
Official Documentation: Claude Code MCP Guide
Edit your configuration file (Settings > Developer > Edit Config):
~/Library/Application Support/Claude/claude_desktop_config.json%APPDATA%/Claude/claude_desktop_config.jsonThis configuration applies globally to all conversations.
Note: Restart Claude Desktop after configuration changes.
Verify: Look for Mandoline tools when you click the "Search and tools" button.
Official Documentation: MCP Quickstart Guide
Create or edit your MCP configuration file:
You can use your global configuration (affects all projects) ~/.cursor/mcp.json or project-local configuration (current project only) .cursor/mcp.json (in project root)
Note: Restart Cursor after configuration changes.
Verify: Check the Output panel (Ctrl+Shift+U) → "MCP Logs" for successful connection, or look for Mandoline tools in the Composer Agent.
Official Documentation: Cursor MCP Guide
Only needed if you want to run the server locally or contribute to development. Most users should use the hosted server above.
Prerequisites: Node.js 18+ and npm
The server runs on http://localhost:8080 by default.
To use your local server instead of the hosted one, replace https://mandoline.ai/mcp with http://localhost:8080/mcp in the client configurations above.
Once integrated, you can use Mandoline evaluation tools directly in your AI assistant conversations.
| Tool | Purpose |
|---|---|
create_metric | Define custom evaluation criteria for your specific tasks |
batch_create_metrics | Create multiple evaluation metrics in one operation |
get_metric | Retrieve details about a specific metric |
get_metrics | Browse your metrics with filtering and pagination |
update_metric | Modify existing metric definitions |
| Tool | Purpose |
|---|---|
create_evaluation | Score prompt/response pairs against your metrics |
batch_create_evaluations | Evaluate the same content against multiple metrics |
get_evaluation | Retrieve evaluation results and scores |
get_evaluations | Browse evaluation history with filtering and pagination |
update_evaluation | Add metadata or context to evaluations |
Access Mandoline's documentation and reference materials directly in your AI assistant, including model comparison guides and evaluation best practices.
Apache-2.0 License - see the LICENSE file for details.
This server cannot be installed
hybrid server
The server is able to function both locally and remotely, depending on the configuration or use case.
Enables AI assistants to reflect on, critique, and continuously improve their performance using Mandoline's evaluation framework. Provides tools for creating custom evaluation metrics and scoring prompt/response pairs to measure AI assistant quality.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/mandoline-ai/mandoline-mcp-server'
If you have feedback or need assistance with the MCP directory API, please join our Discord server