---
title: "Opik MCP Server: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/mcp-servers/opik-mcp/
description: "Reads and scores Opik traces, experiments, datasets and prompts from an MCP client. Comet hosts it for Opik Cloud; self-hosted and open-source Opik run the same package locally over stdio."
---

# Opik MCP Server: review, setup & limitations · Undominated.ai

> Reads and scores Opik traces, experiments, datasets and prompts from an MCP client. Comet hosts it for Opik Cloud; self-hosted and open-source Opik run the same package locally over stdio.

[← Explore all mcp servers](/mcp-servers/)

LLM OBSERVABILITY / Comet

# Opik MCP Server

Reads and scores Opik traces, experiments, datasets and prompts from an MCP client. Comet hosts it for Opik Cloud; self-hosted and open-source Opik run the same package locally over stdio.

 [See setup guidance ↓](#setup)[Original source ↗](https://github.com/comet-ml/opik-mcp)

SOURCE REVIEW

 Reviewed 2026-10-13
 Evidence 4 linked sources
 Publisher Comet
 Licence [Apache-2.0 ↗](https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/LICENSE)
 Revision e73758164ca2
 [Read what was—and wasn’t—checked ↓](#review)

“Read and score Opik traces, experiments, datasets and prompts from your AI assistant.”

 [Comet · upstream description ↗](https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/server.json) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - The README at this commit is published by Comet for its own Opik platform and documents both a hosted Streamable HTTP endpoint (https://www.comet.com/opik/api/v1/mcp, browser OAuth, no API key) and a local stdio package started as uvx opik-mcp.
- The README's write tool lists explicit evaluation operations (score.create, prompt_version.save, dataset.create, experiment.create), and the LICENSE at this commit is Apache-2.0, Copyright (c) Comet ML, Inc, matching pyproject.toml.

### A good fit for

 - Reading failed traces, attaching scores and saving prompt versions in an Opik workspace from a coding agent
- Creating datasets and experiments, or turning on Diagnostics scans for a project, from the same chat session

### Weigh up before choosing

 - The write tool logs traces and spans, attaches scores and comments, saves prompt versions, creates datasets and experiments, and can resolve or close Diagnostics issues. It is not a read-only server.
- The README says the older TypeScript npm package (opik-mcp@2) is deprecated and stops serving requests on 2026-11-15. Its quick-start and check commands use @latest, which is not pinned to this commit.
- The README notes Cursor has a hard 60-second tool-call timeout. The server was not installed and no Opik workspace was contacted.

02 / THE REVIEW RECORD

## What we actually inspected.

Source review has boundaries. A clear record is more useful than a “safe” badge.

### Material inspected

 - README.md at e73758164ca282c1e7ac37934923563ed98479d8
- LICENSE at the same commit
- pyproject.toml at the same commit
- server.json at the same commit

### Our findings

 - The README says an Opik API key is not validated locally: it is forwarded verbatim to opik-backend, while an OAuth access token is checked on every HTTP request against the backend's token introspection.
- For local open-source Opik the README warns the backend does not authenticate requests, so an HTTP opik-mcp in front of it is as open as that REST API; it recommends stdio and the default 127.0.0.1 bind.
- pyproject.toml names opik-mcp, requires Python >=3.13, and declares the opik-mcp console script.

### Not established by this review

 - The uvx opik@latest mcp configure installer
- The hosted OAuth sign-in
- Any read or write against an Opik workspace

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

## Connect a server deliberately.

[Upstream setup instructions ↗](https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/README.md)

DOCUMENTED COMMAND

 uvx opik@latest mcp configure Copy command ↗

Copying does not execute this command. It may retrieve a newer version than the reviewed source.

 - The README's quick start above registers the server in supported clients, installs the Opik skill pack and checks the connection. It was not run.
- For Opik Cloud, adding https://www.comet.com/opik/api/v1/mcp as an HTTP server and signing in in the browser is enough. For self-hosted or open-source Opik, register uvx opik-mcp with OPIK_URL, or COMET_URL_OVERRIDE, OPIK_WORKSPACE and OPIK_API_KEY. Do not commit the key.

### Before you start

 - uv
- An Opik Cloud account, a self-hosted Comet platform or an open-source Opik deployment

### Documented configuration example

Use the client and authentication context described in the setup guidance. Configuration formats vary between hosts.

 {
 "mcpServers": {
 "opik-mcp": {
 "type": "stdio",
 "command": "uvx",
 "args": [
 "opik-mcp"
 ],
 "env": {
 "OPIK_URL": "http://localhost:5173/api"
 }
 }
 }
} Copy configuration ↗

### Compatibility

Claude Code · Codex · Cursor · VS Code Copilot · OpenCode

### Implementation

Python

### Transports

stdio · Streamable HTTP

### Hosted OAuth endpoint or local process with Opik settings

 - Reads traces, spans, experiments, datasets and prompts in the signed-in Opik workspace
- Writes traces, scores, comments, prompt versions, datasets and experiments through the write tool

### Cost model

Apache-2.0 server source. Opik Cloud or self-hosted deployment terms apply; no price was reviewed.

04 / FOLLOW THE EVIDENCE

## The source trail.

Our notes are separate from the original resource. Check upstream before adopting a new version.

 - [Opik MCP README ↗](https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/README.md) Checked 2026-10-13 https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/README.md Supports: Hosted and local servers, Tools and write operations, Deprecation of the npm package, Transport security notes
- [Opik MCP LICENSE ↗](https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/LICENSE) Checked 2026-10-13 https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/LICENSE Supports: Apache License 2.0, Copyright (c) Comet ML, Inc
- [Opik MCP pyproject.toml ↗](https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/pyproject.toml) Checked 2026-10-13 https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/pyproject.toml Supports: Package name opik-mcp, Python >=3.13, License Apache-2.0
- [Opik MCP server.json ↗](https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/server.json) Checked 2026-10-13 https://github.com/comet-ml/opik-mcp/blob/e73758164ca282c1e7ac37934923563ed98479d8/server.json Supports: Description, Remote endpoint and PyPI package identity

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### Helicone MCP Server ↗ Connects an assistant to Helicone request logs and sessions, and can send new LLM requests through the Helicone AI Gateway. It runs locally over stdio via the published npm package with a Helicone API key.](/mcp-servers/helicone/)[### MLflow MCP Server ↗ Ships inside the mlflow package and exposes trace search, feedback and assessment logging, and optionally experiment, run, model and deployment commands, to an MCP client.](/mcp-servers/mlflow-mcp/)[### OpenTelemetry MCP Server ↗ Queries LLM application traces in Jaeger, Grafana Tempo or Traceloop, with tools that aggregate token usage by model and find slow, token-heavy or failing traces.](/mcp-servers/opentelemetry-mcp/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Inspect resource evidence](/tools/)
- [Choose a host](/agents/)
- [Inspect reusable workflows](/skills/)
- [Use Undominated data](/api/)
