Awesome MCP › Software Development

comet-ml/opik-mcp

⭐ 218 TypeScript added to this list on 2025-05-06 repository created 2025-03-11

Opik MCP Server is the official Model Context Protocol server for Opik, the open source LLM observability and evaluation platform from Comet. It connects an AI host such as Claude Code, Cursor, VS Code Copilot or the MCP Inspector to an Opik workspace so that traces, spans, experiments, test suites and prompts can be read and written from a chat session. The server is distributed as a Python package run through uvx and authenticates with an OPIK_API_KEY plus an optional OPIK_WORKSPACE name, defaulting to the workspace called default, which is correct for local and open source installs. Self hosted deployments point at their own instance with COMET_URL_OVERRIDE. An earlier TypeScript implementation is deprecated, with a documented migration path and a stated sunset date. The tool surface is deliberately small and outcome oriented. The read tool fetches a single entity by UUID, by name or by an opik:// URI and inlines children so composite reads return the full picture. The list tool browses a collection with an optional name filter and pagination, requiring a parent UUID for project scoped types. The write tool logs traces and spans, records scores and comments, saves prompt versions and manages test suites and experiments, while the schema tool lets the model introspect write payload shapes before constructing them. The ask_ollie tool forwards investigative or cross entity questions to Ollie, the in product Opik assistant, and returns a thread id that preserves context on follow ups; run_experiment drives an evaluation experiment end to end. Both of those are available on Comet Cloud only. Writes that Ollie performs execute without per action confirmation by default, each auto approval is recorded as a JSON audit row on a dedicated logger, and setting OPIK_MCP_AUTO_APPROVE to disabled requires explicit confirmation instead. The documentation also notes host limits such as the fixed sixty second tool call timeout in Cursor. It targets LLM engineers who already run Opik.

https://github.com/comet-ml/opik-mcp

mcp-serveropikcometllm-observabilitytracingevaluationprompt-managementtypescriptpython