# Arize Phoenix

An open-source AI observability and evaluation platform for tracing, evaluating and troubleshooting LLM applications, with an integrated debugging agent and a remote MCP server for querying traces.

Arize Phoenix is Arize's open-source AI observability platform for experimenting with, evaluating and troubleshooting LLM applications; a separate managed offering covers production workflows using the same underlying instrumentation. Phoenix provides OpenTelemetry-based tracing of an application's runtime, LLM-based evaluation of response and retrieval quality, versioned datasets of examples for experimentation and fine-tuning, and experiment tracking for changes to prompts, models and retrieval.

A playground lets a developer optimize prompts, compare models, adjust parameters and replay traced calls, and prompt management supports version control, tagging and systematic testing of prompt changes. An integrated engineering agent helps debug traces and iterate on prompts, and a remote MCP server lets Claude Code, Cursor and other MCP clients query an instance's traces, datasets and experiments directly.

It is vendor- and language-agnostic, with built-in support for popular agent frameworks and LLM providers, and installs with pip or conda, or runs without installation through uvx, on a local machine, in a container, or in the cloud.

Website: https://arize.com/docs/phoenix