Arize Phoenix
A Claude Code plugin for the Phoenix LLM observability platform: remote MCP plus skills for traces, evals and instrumentation
Install
claude plugin marketplace add Arize-ai/phoenix
claude plugin install arize-phoenix@arize-phoenixRun /mcp afterwards and complete the Phoenix login.
This is third-party code. Review the repository files before installing.
What it does
Phoenix is an open-source, self-hosted platform for tracing and evaluating LLM apps. The plugin connects Claude Code to your Phoenix instance through its remote MCP server and adds three skills. phoenix-cli teaches the agent to fetch traces, analyze errors, inspect datasets and experiments and query GraphQL with the px CLI. phoenix-evals helps build and run evaluators, and phoenix-tracing instruments apps with OpenInference conventions. The CLI setup command detects your framework and provider and wires up trace export.
Who it is for. For developers of LLM apps and agents who debug and evaluate them in Phoenix.
Good fit when
- You want the agent to analyze failing LLM app traces itself
- You need to add OpenInference tracing to an existing project
- You need to build evaluators and run an experiment on a dataset
Not a fit when
- You do not use Phoenix and do not plan to deploy it
- You need observability for regular web services rather than LLMs
Example request
Find error traces in Phoenix from the last hour and explain what is breakingLimitations
Requires a deployed Phoenix 19.0.0 or later, defaulting to localhost:6006. The MCP server signs in via browser OAuth; for headless use register the server manually with an API key. The px CLI reads PHOENIX_ENDPOINT and PHOENIX_API_KEY from the environment.
How to disable. Run claude plugin uninstall arize-phoenix@arize-phoenix or disable the plugin via /plugin.
Security check
- Reads traces that may contain user data
- The tracing skill changes project code and dependencies
README in short
The README presents Phoenix as an open observability platform: OpenTelemetry tracing, evals, datasets, experiments, a prompt playground and a built-in agent. Run it locally via pip or uvx, deploy with Docker and Helm. For coding agents it offers three surfaces: CLI, skills and a remote MCP server. The plugin README covers install, endpoint configuration and OAuth sign-in.
FAQ
How do I set my server address?
Claude Code asks when enabling the plugin; you can also pass --config endpoint=... on install or change it with /plugin configure.
Can I use it without the plugin?
Yes, run the CLI with npx @arizeai/phoenix-cli and point any MCP client at your instance's /mcp endpoint.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Anthropic's coding agent for the terminal, IDE and GitHub: understands your codebase, runs tasks and handles git