Opik MCP Server
平台与服务by comet-ml
通过 Model Context Protocol 与 Opik 的 prompts、traces 和 metrics 交互,便于统一访问与分析。
通过 MCP 把 Opik 的提示词、追踪和指标统一打通,少了来回切换与数据割裂,做 AI 应用分析调试更省心。
什么是 Opik MCP Server?
通过 Model Context Protocol 与 Opik 的 prompts、traces 和 metrics 交互,便于统一访问与分析。
README
<a href="https://www.comet.com/docs/opik/prompt_engineering/mcp_server"><img src="https://badge.mcpx.dev?status=on" title="MCP Enabled" alt="MCP Enabled" /></a>
<a href="https://doi.org/10.5281/zenodo.15411156"><img src="https://zenodo.org/badge/DOI/10.5281/zenodo.15411156.svg" alt="DOI" /></a>
[!IMPORTANT] This repository ships the MCP server implementation only. We do not currently provide a hosted remote MCP service for Opik. If you run
streamable-httpremotely, authentication is fail-closed by default.
Why this server
Opik MCP Server gives MCP-compatible clients one interface for:
- Prompt lifecycle management
- Workspace, project, and trace exploration
- Metrics and dataset operations
- MCP resources and resource templates for metadata-aware flows
Quickstart
1. Run with npx
# Opik Cloud
npx -y opik-mcp --apiKey YOUR_API_KEY
For self-hosted Opik, pass --apiUrl (for example http://localhost:5173/api) and use your local auth strategy.
2. Add to your MCP client
Cursor (.cursor/mcp.json):
{
"mcpServers": {
"opik": {
"command": "npx",
"args": ["-y", "opik-mcp", "--apiKey", "YOUR_API_KEY"]
}
}
}
VS Code / GitHub Copilot (.vscode/mcp.json):
{
"inputs": [
{
"type": "promptString",
"id": "opik-api-key",
"description": "Opik API Key",
"password": true
}
],
"servers": {
"opik-mcp": {
"type": "stdio",
"command": "npx",
"args": ["-y", "opik-mcp", "--apiKey", "${input:opik-api-key}"]
}
}
}
Windsurf (raw config):
{
"mcpServers": {
"opik": {
"command": "npx",
"args": ["-y", "opik-mcp", "--apiKey", "YOUR_API_KEY"]
}
}
}
More client-specific examples: docs/ide-integration.md
Run from source
git clone https://github.com/comet-ml/opik-mcp.git
cd opik-mcp
npm install
npm run build
Optional local config:
cp .env.example .env
Start the server:
npm run start:stdio
npm run start:http
Transport modes
| Transport | Use case | Command |
|---|---|---|
stdio | Local MCP integration (same machine as client) | npm run start:stdio |
streamable-http | Remote/self-hosted MCP endpoint (/mcp) | npm run start:http |
Remote auth defaults (streamable-http)
Authorization: Bearer <OPIK_API_KEY>orx-api-keyis required by default.- Workspace is resolved server-side (token map recommended); workspace headers are not trusted by default.
- In remote mode, request-context workspace takes precedence over tool
workspaceName. - Missing or invalid auth returns HTTP
401.
Key environment flags:
STREAMABLE_HTTP_REQUIRE_AUTH(defaulttrue)STREAMABLE_HTTP_VALIDATE_REMOTE_AUTH(defaulttrue, except test env)REMOTE_TOKEN_WORKSPACE_MAP(JSON token-to-workspace map)STREAMABLE_HTTP_TRUST_WORKSPACE_HEADERS(defaultfalse)
Deep dive: docs/streamable-http-transport.md
Toolsets
Toolsets let you narrow which capabilities are enabled:
coreintegrationexpert-promptsexpert-datasetsexpert-trace-actionsexpert-project-actionsmetricsall(enables all modern toolsets)
Configure via:
- CLI:
--toolsets all - Env:
OPIK_TOOLSETS=core,expert-prompts,metrics
Details: docs/configuration.md
MCP resources and prompts
resources/listexposes static URIs (for exampleopik://workspace-info)resources/templates/listexposes dynamic URI templates (for exampleopik://projects/{page}/{size})resources/readsupports static and templated URIsprompts/listandprompts/getexpose workflow prompts
Development
# Lint
npm run lint
# Test
npm test
# Build
npm run build
# Run precommit checks
make precommit
Documentation
Contributing
Please read CONTRIBUTING.md before opening a PR.
Citation
If you use this project in research, cite:
Comet ML, Inc, Koc, V., & Boiko, Y. (2025). Opik MCP Server. Github. https://doi.org/10.5281/zenodo.15411156
BibTeX:
@software{CometML_Opik_MCP_Server_2025,
author = {{Comet ML, Inc} and Koc, V. and Boiko, Y.},
title = {{Opik MCP Server}},
year = {2025},
publisher = {GitHub},
url = {https://doi.org/10.5281/zenodo.15411156},
doi = {10.5281/zenodo.15411156}
}
Citation metadata is also available in CITATION.cff.
License
Apache 2.0
常见问题
Opik MCP Server 是什么?
通过 Model Context Protocol 与 Opik 的 prompts、traces 和 metrics 交互,便于统一访问与分析。
相关 Skills
Slack动图
by anthropics
面向Slack的动图制作Skill,内置emoji/消息GIF的尺寸、帧率和色彩约束、校验与优化流程,适合把创意或上传图片快速做成可直接发送的Slack动画。
✎ 帮你快速做出适配 Slack 的动图,内置约束规则和校验工具,少踩上传与播放坑,做表情包和演示都更省心。
MCP构建
by anthropics
聚焦高质量 MCP Server 开发,覆盖协议研究、工具设计、错误处理与传输选型,适合用 FastMCP 或 MCP SDK 对接外部 API、封装服务能力。
✎ 想让 LLM 稳定调用外部 API,就用 MCP构建:从 Python 到 Node 都有成熟指引,帮你更快做出高质量 MCP 服务器。
接口测试套件
by alirezarezvani
扫描 Next.js、Express、FastAPI、Django REST 的 API 路由,自动生成覆盖鉴权、参数校验、错误码、分页、上传与限流场景的 Vitest 或 Pytest 测试套件。
✎ 帮你把API与集成测试自动化跑顺,减少回归漏测;能力全面,尤其适合复杂接口场景的QA团队。
相关 MCP Server
Slack 消息
编辑精选by Anthropic
Slack 是让 AI 助手直接读写你的 Slack 频道和消息的 MCP 服务器。
✎ 这个服务器解决了团队协作中需要 AI 实时获取 Slack 信息的痛点,特别适合开发团队让 Claude 帮忙汇总频道讨论或发送通知。不过,它目前只是参考实现,文档有限,不建议在生产环境直接使用——更适合开发者学习 MCP 如何集成第三方服务。
by netdata
io.github.netdata/mcp-server 是让 AI 助手实时监控服务器指标和日志的 MCP 服务器。
✎ 这个工具解决了运维人员需要手动检查系统状态的痛点,最适合 DevOps 团队让 Claude 自动分析性能数据。不过,它依赖 NetData 的现有部署,如果你没用过这个监控平台,得先花时间配置。
by d4vinci
Scrapling MCP Server 是专为现代网页设计的智能爬虫工具,支持绕过 Cloudflare 等反爬机制。
✎ 这个工具解决了爬取动态网页和反爬网站时的头疼问题,特别适合需要批量采集电商价格或新闻数据的开发者。不过,它依赖外部浏览器引擎,资源消耗较大,不适合轻量级任务。