Open-source Model Context Protocol (MCP) server for the Comet Opik platform. Offers stdio and experimental SSE transports, giving IDEs (e.g., Cursor) a unified interface to manage prompts, projects, traces and metrics.
https://github.com/comet-ml/opik-mcpYou're building with LLMs, but every time you want to check how your prompts are performing or analyze traces, you're alt-tabbing to a web dashboard. That friction adds up—and it breaks your flow.
The Opik MCP Server changes this. It brings Comet's LLM evaluation and monitoring platform directly into your IDE through the Model Context Protocol, so you can manage prompts, track traces, and analyze metrics without ever leaving your editor.
When you're iterating on LLM applications, you need constant visibility into what's happening:
But most LLM observability tools force you out of your development environment. You write code, run tests, then jump to a web interface to check results. This context switching kills productivity and makes it harder to iterate quickly.
The Opik MCP Server bridges this gap by implementing the Model Context Protocol. Once configured, you get:
Prompt Management in Your Editor: Create, update, and version prompts directly from Cursor or other MCP-compatible IDEs. No more copying prompts back and forth between your code and a web interface.
Real-Time Trace Analysis: See LLM call traces and responses inline with your code. Debug issues immediately instead of hunting through logs later.
Metrics at Your Fingertips: Check evaluation scores, response times, and other metrics right where you're working. Make data-driven decisions without breaking focus.
Project Organization: Manage multiple LLM projects and workspaces through a unified interface that integrates with your existing development workflow.
Prompt Engineering Iterations: You're refining a summarization prompt. Instead of running your script, then opening a browser to check Opik's dashboard, you see the results and metrics directly in your IDE. Adjust the prompt, test again, and compare results—all in one place.
Model Comparison: Testing GPT-4 vs Claude for your use case? The MCP server lets you run experiments and compare performance metrics without leaving your development environment. You can see which model performs better on your specific tasks while you're writing the integration code.
Production Debugging: When something goes wrong with your LLM application, you need to trace the issue quickly. With Opik's traces available in your IDE, you can analyze the problematic requests alongside your code, making it easier to identify and fix issues.
Team Collaboration: Your team can share prompts and evaluation results through Opik's centralized platform, but developers can access everything through their preferred IDE. No need to standardize on a specific web interface.
The server supports both stdio (for local development) and Server-Sent Events (for remote access), giving you flexibility in how you integrate it:
Local Development: Use stdio transport for tight integration with your local development environment. Perfect for individual developers who want zero-latency access to their LLM data.
Team Environments: The experimental SSE transport enables remote access and multiple simultaneous connections, useful for team setups or when running the server on a different machine.
The setup is straightforward for any Node.js developer:
git clone https://github.com/comet-ml/opik-mcp.git
cd opik-mcp
npm install && npm run build
For Cursor integration, drop a simple JSON config in your .cursor/mcp.json:
{
"mcpServers": {
"opik": {
"command": "/path/to/node",
"args": ["/path/to/opik-mcp/build/index.js", "--apiUrl", "https://www.comet.com/opik/api", "--apiKey", "YOUR_API_KEY"]
}
}
}
The server connects to your existing Opik workspace, so you keep all your historical data and team configurations.
LLM development is inherently iterative. You need tight feedback loops to understand what's working and what isn't. When your observability tools are separate from your development environment, you lose that tight coupling.
The Opik MCP Server doesn't just save you time—it changes how you approach LLM development. Instead of treating monitoring as a separate phase, it becomes part of your natural development flow. You write code, see results, adjust, and iterate—all without friction.
If you're serious about building reliable LLM applications, the context switching tax of separate monitoring tools is holding you back. This MCP server eliminates that friction and keeps you in the zone where you do your best work.