Arize Phoenix MCP Server: Setup for Claude, Cursor and Other Apps

• Lists the projects in your Phoenix

  Returns the projects that organize your traces and other data.

  Try: “What projects do I have in Phoenix?”

  Network · list-projects

• Shows one project

  Returns details for a single project.

  Try: “Show me the details of project 'checkout-agent'.”

  Network · get-project

• Lists recent traces in a project

  Lists traces for a project so you can find slow or failing runs.

  Try: “Which traces in 'checkout-agent' failed today?”

  Network · list-traces

• Shows one trace with its spans

  Returns a single trace so the assistant can walk through what happened.

  Try: “Walk me through trace 4f2a and tell me where it went wrong.”

  Network · get-trace

• Fetches spans from a project

  Returns spans from a project, optionally narrowed by your filters.

  Try: “Show LLM spans from the last hour that took over 10 seconds.”

  Network · get-spans

• Shows the scores and notes on spans

  Returns annotations attached to spans, such as evaluation scores and human feedback.

  Try: “What annotations are on these spans?”

  Network · get-span-annotations

• Lists conversation sessions in a project

  Lists the sessions in a project, which group the traces of one conversation.

  Try: “List the latest sessions in 'support-bot'.”

  Network · list-sessions

• Shows one session

  Returns a session with its conversation flow and session-level annotations.

  Try: “Show me session 9c1 from start to finish.”

  Network · get-session

• Lists the annotation types you have set up

  Lists the annotation configs, which define the scores and labels you can attach.

  Try: “What annotation types does my Phoenix use?”

  Network · list-annotation-configs

• Lists the prompts you manage in Phoenix

  Returns your saved prompts.

  Try: “List my Phoenix prompts.”

  Network · list-prompts

• Shows one prompt

  Returns a prompt by name.

  Try: “Show me the 'support-reply' prompt.”

  Network · get-prompt

• Shows the newest version of a prompt

  Returns the most recent version of a prompt.

  Try: “What is the latest version of 'support-reply'?”

  Network · get-latest-prompt

• Finds a prompt by its identifier

  Returns a prompt using its identifier.

  Try: “Get the prompt with this identifier.”

  Network · get-prompt-by-identifier

• Shows one prompt version

  Returns a specific version of a prompt.

  Try: “Show version 3 of 'support-reply'.”

  Network · get-prompt-version

• Lists all versions of a prompt

  Returns the version history of a prompt.

  Try: “How has 'support-reply' changed over time?”

  Network · list-prompt-versions

• Finds the prompt version that carries a tag

  Returns the prompt version with a given tag, such as production.

  Try: “Which version of 'support-reply' is tagged production?”

  Network · get-prompt-version-by-tag

• Lists the tags on a prompt version

  Returns the tags attached to a prompt version.

  Try: “What tags does this prompt version have?”

  Network · list-prompt-version-tags

• Adds a tag to a prompt version

  Tags a prompt version, for example to mark it as production.

  Try: “Tag version 4 of 'support-reply' as staging.”

  Write · add-prompt-version-tag

• Creates a prompt or saves a new version of it

  Creates a prompt, or adds a new version when it exists. You set the template, model provider, model name and temperature.

  Try: “Save a new version of 'support-reply' with a friendlier tone.”

  Write · upsert-prompt

• Lists your datasets

  Returns the datasets in your Phoenix.

  Try: “What datasets do I have?”

  Network · list-datasets

• Shows one dataset

  Returns details of a dataset.

  Try: “Show me the 'refund-questions' dataset.”

  Network · get-dataset

• Shows the examples in a dataset

  Returns the input and output examples stored in a dataset.

  Try: “Show me 10 examples from 'refund-questions'.”

  Network · get-dataset-examples

• Lists the experiments run on a dataset

  Returns experiments linked to a dataset.

  Try: “Which experiments ran on 'refund-questions'?”

  Network · get-dataset-experiments

• Adds new examples to a dataset

  Adds input and output examples, with optional metadata, to a dataset in Phoenix.

  Try: “Add these five tricky questions to 'refund-questions'.”

  Write · add-dataset-examples

• Lists experiments for a dataset

  Returns the experiments that were run against a dataset.

  Try: “List the experiments for 'refund-questions'.”

  Network · list-experiments-for-dataset

• Shows the results of one experiment

  Returns an experiment's results so the assistant can compare them.

  Try: “Summarize the results of experiment 12.”

  Network · get-experiment-by-id

• Answers questions about Phoenix and OpenInference using Arize's support assistant

  Sends your question to a hosted RunLLM support service at mcp.runllm.com and returns its answer. It covers tracing, datasets, experiments, prompts, evals and annotations. Your question leaves your machine and goes to that service.

  Try: “How do I send traces from my LangChain app to Phoenix?”

  Network · phoenix-support

Watch out for

• It can change prompts and datasets

  Three tools save prompt versions, tag them and add dataset examples. There is no read-only switch, so use a key that suits that. github.com

• The support tool sends your question to a third party

  phoenix-support forwards the question to a RunLLM service. Do not put private data in questions to it. github.com

• Traces may hold sensitive prompts and outputs

  Trace and span tools pass whatever your app logged to the assistant. Check what your traces contain first. github.com

Security advisories