Skip to main content
Maintenance mode. @arizeai/phoenix-mcp continues to receive bug fixes, but new capabilities land in the Remote MCP Server built into the Phoenix server. Reach for this package only when your Phoenix version doesn’t serve the /mcp endpoint. New to Phoenix MCP? Start with the MCP Overview.
The Phoenix MCP Server (@arizeai/phoenix-mcp) is a local stdio server, launched with npx, that connects AI assistants directly to your Phoenix instance for managing:
  • Projects, Traces, and Spans: Explore recent traces, inspect spans, and analyze annotations
  • Sessions: Review conversation flows and session annotations
  • Annotation Configs: Inspect the available labeling and scoring configs in Phoenix
  • Prompts Management: Create, list, update, and iterate on prompts
  • Datasets: Explore datasets and synthesize new examples
  • Experiments: Pull experiment results and visualize them with the help of an LLM

​
Connecting the Phoenix MCP Server

The package runs as a local stdio process. In every client, pass your Phoenix endpoint with --baseUrl and an API key with --apiKey (for a local instance, --baseUrl http://localhost:6006).
1

Add the Phoenix MCP server

Restart your Claude Code session to start using the Phoenix MCP tools.
1

Add the Phoenix MCP server

In Settings → Developer → Edit Config, add:
Save the file and relaunch Claude Desktop.
1

Add the Phoenix MCP server

Add to ~/.cursor/mcp.json (or project .cursor/mcp.json):
Any client that supports stdio MCP servers works. Configure it to run npx -y @arizeai/phoenix-mcp@latest --baseUrl <your-phoenix-url> --apiKey <your-api-key>.

​
Using the Phoenix MCP Server

The MCP server can be used to interact with projects, traces, spans, sessions, annotation configs, prompts, experiments, and datasets. It can retrieve operational data, inspect prompt and experiment artifacts, and perform the existing prompt and dataset write flows. Some good questions to try:
  1. Show me the latest traces in my default Phoenix project
  2. Show me the last 10 sessions in my support-agent project
  3. What annotation configs do I have in Phoenix?
  4. What prompts do I have in Phoenix?
  5. Create a new prompt in Phoenix that classifies user intent
  6. Summarize the Phoenix experiments run on my agent inputs dataset
  7. Visualize the results of my jailbreak dataset experiments in Phoenix

​
Hoping to see additional functionality?

@arizeai/phoenix-mcp is open-source! Issues and PRs welcome.

phoenix-mcp