# Opik MCP
> Comet's official MCP integration for Opik lets an AI client inspect LLM traces and projects, record scores and comments, and manage prompts, datasets, and experiments in the connected Opik workspace.
Canonical: https://www.mcpfolio.com/servers/opik-mcp
Last updated: 2026-10-28
## Catalog record

- Publisher: Comet (official Opik project)
- Categories: Uncategorized

## Website

https://www.comet.com/site/products/opik/

## Repository

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

## Transports

Streamable HTTP for hosted Opik Cloud; stdio for the local package.

## Authentication

Opik Cloud hosted endpoint uses browser OAuth; no API key or manual workspace setting is needed. Local open-source Opik generally needs no key unless authentication is enabled. Self-hosted Comet or local-server-to-Cloud setups may require OPIK_API_KEY and OPIK_WORKSPACE.

## Tools

- **read**: Read an Opik entity by ID, name, or Opik URI.
- **list**: List and filter Opik entities with pagination.
- **write**: Create or update traces, spans, scores, comments, prompts, datasets, experiments, and selected agent-insights records.
- **schema**: Inspect the required shape, scope, and example for supported operations without calling the backend.

## Sources

- [official_repository_documentation](https://github.com/comet-ml/opik-mcp)
- [official_package_metadata](https://github.com/comet-ml/opik-mcp/blob/main/pyproject.toml)
- [verified_publisher_organization_avatar](https://github.com/comet-ml.png?size=300)
- [discovery_directory](https://awesome-mcp.tools/)
- [official_or_maintainer_tool_reference](https://github.com/comet-ml/opik-mcp)

## Setup

Use the interactive setup panel on [the detail page](https://www.mcpfolio.com/servers/opik-mcp) to view instructions for each supported client.