Opik MCP Server

@comet-ml

Basic information

About Opik MCP Server

Model Context Protocol (MCP) implementation for Opik enabling seamless IDE integration and unified access to prompts, projects, traces, and metrics.

#typescript #generative-ai #mcp-server #modelcontextprotocol

Opik MCP Server config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "opik-mcp": {
      "command": "npx",
      "args": [
        "@modelcontextprotocol/inspector",
        "uvx",
        "opik-mcp"
      ]
    }
  }
}

Opik MCP Server tools

Tool list not available yet

This listing does not currently have a tool list to display. This does not mean the server has no tools; check its documentation for supported capabilities.

Fetch the live tool list by running this server in a temporary sandbox using the button above.

Overview

What is Opik Mcp?

Opik Mcp is a Model Context Protocol server that connects AI hosts (Claude Code, Cursor, VS Code Copilot, MCP Inspector) directly to an Opik workspace. It lets LLM engineers read traces, log scores, save prompt versions, and ask the Opik in-product assistant Ollie investigative questions — all from the chat interface of their AI coding tool.

How to use Opik Mcp?

Install the Python package (requires Python 3.13+) via uvx opik-mcp. Configure the OPIK_API_KEY and optionally OPIK_WORKSPACE environment variables, then add the server to your MCP client config (e.g., ~/.claude.json for Claude Code, .cursor/mcp.json for Cursor, or .vscode/mcp.json for VS Code Copilot). Restart your host and ask, for example: "list my Opik projects."

Key features of Opik Mcp

Six outcome-oriented tools: read, list, ask_ollie, write, schema, run_experiment
Universal read and list by id, name, or opik:// URI
In-workspace investigation via Ollie with YOLO-mode auto-approval
Write operations: log traces, spans, scores, comments, prompt versions, test suites, experiments
Schema introspection for constructing valid write payloads
End-to-end experiment execution via Ollie on Comet Cloud

Use cases of Opik Mcp

Investigate why an experiment regressed on factuality by asking Ollie in chat
Score a trace on helpfulness directly from the AI assistant without leaving the editor
Browse and filter Opik projects, experiments, and traces by name or parent
Save new prompt versions and manage test suites through conversational commands
Run evaluation experiments end-to-end by sending an experiment_config dict

Frequently asked questions

What does Opik Mcp require to run?

Python 3.13+, the `uv` tool, an `OPIK_API_KEY` from your Opik workspace, and optionally an `OPIK_WORKSPACE` name (defaults to `default` for local/OSS installs).

Where does Opik Mcp store data?

All data lives in your Opik workspace — either Comet Cloud (`www.comet.com`) or a self-hosted instance configured via `COMET_URL_OVERRIDE`. Opik Mcp is a stateless proxy; it doesn't persist data locally.

Are there known host limits?

Cursor enforces a 60-second tool-call timeout that doesn't reset on progress notifications, so long `ask_ollie` turns will fail on Cursor.

Which transports does Opik Mcp support?

It defaults to `stdio` (host-launched) and supports `streamable-http` to listen on a port. Set `OPIK_MCP_TRANSPORT=streamable-http` and configure `OPIK_MCP_HOST`/`OPIK_MCP_PORT` as needed.

Is `ask_ollie` or `run_experiment` available on self-hosted Opik?

No. Both `ask_ollie` and `run_experiment` are available on Comet Cloud only. On self-hosted installs, use `read`, `list`, `schema`, and `write` directly.

Comments

More Developer Tools MCP servers

LocalCan

Gives AI agents public URLs (tunnels) for localhost, live HTTP traffic inspection, snapshot publishing, and access control. Part of LocalCan, the ngrok alternative for Mac, Windows and Linux. Free plan.

Cronwatch

cronwatchdev

CronWatch records every run of your scheduled jobs in your own database, and speaks up when one is missed, fails, gets stuck, runs slow or costs too much. CronWatch is free and open-source.

Mcp-Warden

ermoluk

Local-first credential broker for MCP. AI coding agents (Claude Code, Cursor, Codex…) call APIs, SSH and databases through scoped presets; secrets stay in your OS keychain and never enter the agent's context.

Eddyter AI Rich Text Editor MCP

EddyterAI

Set up and manage the Eddyter AI rich text editor by talking to your AI agent — for Claude, Cursor, Codex and other MCP tools. A Model Context Protocol server that lets AI tools add Eddyter to a React, Next.js, Vite or R

TaskerArmy Agent

TaskerArmy

Ask Claude or ChatGPT what Shopify theme optimization tasks are pending on your store, a remote MCP server for TaskerArmy Agent accounts.