---
title: "Phoenix Remote MCP: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/mcp-servers/phoenix/
description: "Beta MCP endpoint built into Phoenix 19 or later. The root licence forbids offering Phoenix as a hosted service."
---

# Phoenix Remote MCP: review, setup & limitations · Undominated.ai

> Beta MCP endpoint built into Phoenix 19 or later. The root licence forbids offering Phoenix as a hosted service.

[← Explore all mcp servers](/mcp-servers/)

TRACE REVIEW / Arize AI

# Phoenix Remote MCP

Beta MCP endpoint built into Phoenix 19 or later. The root licence forbids offering Phoenix as a hosted service.

 [See setup guidance ↓](#setup)[Original source ↗](https://github.com/Arize-ai/phoenix)

SOURCE REVIEW

 Reviewed 2026-10-14
 Evidence 2 linked sources
 Publisher Arize AI
 Licence [Elastic License 2.0 ↗](https://github.com/Arize-ai/phoenix/blob/91ea995e0a8db1ac8b67aa2aced89ca29c722c35/LICENSE)
 Revision 91ea995e0a8d
 [Read what was—and wasn’t—checked ↓](#review)

“Connect AI assistants directly to your Phoenix instance through the MCP endpoint built into the Phoenix server …”

 [Arize AI · upstream description ↗](https://github.com/Arize-ai/phoenix/blob/91ea995e0a8db1ac8b67aa2aced89ca29c722c35/docs/phoenix/integrations/remote-mcp.mdx) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - The publisher documents an MCP endpoint on the Phoenix server for searching projects, traces, datasets, experiments, prompts and annotations.
- The same page says the endpoint requires Phoenix 19.0.0 or later and that the older npm package is in maintenance mode.

### A good fit for

 - Compare experiment scores with trace-level failure evidence in a selected Phoenix project.
- Discover API operations and inspect their schemas before an authorized dataset, prompt or annotation change.

### Weigh up before choosing

 - Elastic License 2.0 forbids providing the software to third parties as a hosted or managed service. This catalogue does not redistribute Phoenix and does not offer it as a service.
- The page marks the remote server beta. execute runs agent-written Python in the Monty sandbox unless PHOENIX_ENABLE_MCP_CODE_MODE=false. The server was not run.
- Disabling MCP code mode removes execute but does not disable Monty code evaluators; PHOENIX_ALLOWED_SANDBOX_PROVIDERS governs evaluator providers separately.
- Custom PHOENIX_SKILLS_PATHS resolve relative paths from the server working directory. Only SKILL.md and references are exposed; scripts/assets are not read by this skill-loading feature.

02 / THE REVIEW RECORD

## What we actually inspected.

Source review has boundaries. A clear record is more useful than a “safe” badge.

### Material inspected

 - docs/phoenix/integrations/remote-mcp.mdx
- LICENSE
- 2026-10-14 reread of retained pinned setup/configuration evidence for the specific operational details added in this batch.

### Our findings

 - The licence text opened at this commit is Elastic License 2.0 and includes the hosted-or-managed-service limitation.
- The doc’s default code-mode surface includes execute. The page says filesystem, network and subprocess access are blocked inside Monty, with a 30-second code cap.
- Pinned guide distinguishes OAuth audience scope, server-side master/code-mode switches and the separate evaluator control. The publisher recommends px CLI for most coding-agent workflows; the remote endpoint is useful for ad hoc IDE access.

### Not established by this review

 - Phoenix was not installed. No login, trace query or execute call was made. The Apache-2.0 npm package was not reviewed beyond the maintenance sentence in this doc.
- No authenticated session, installation, permission bypass or package-to-source parity test was performed.

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

## Connect a server deliberately.

[Upstream setup instructions ↗](https://github.com/Arize-ai/phoenix/blob/91ea995e0a8db1ac8b67aa2aced89ca29c722c35/docs/phoenix/integrations/remote-mcp.mdx)

DOCUMENTED COMMAND

 claude mcp add --transport http phoenix http://localhost:6006/mcp Copy command ↗

Copying does not execute this command. It may retrieve a newer version than the reviewed source.

 - Run a Phoenix instance you are allowed to operate, at version 19.0.0 or later.
- The command above is the page's Claude Code line for a local Phoenix at http://localhost:6006; replace that base URL with your instance and log in through /mcp in Claude Code. Do not treat this catalogue entry as a hosted Phoenix service. It was not run, and this review did not connect.

### Before you start

 - A Phoenix 19 or later instance you operate
- OAuth or a Phoenix API key if that instance requires authentication

### Documented configuration example

Use the client and authentication context described in the setup guidance. Configuration formats vary between hosts.

 {
 "mcpServers": {
 "phoenix": {
 "url": "http://localhost:6006/mcp"
 }
 }
} Copy configuration ↗

### Compatibility

MCP clients that can reach a Phoenix /mcp endpoint · Claude Code example in the reviewed page

### Implementation

Python

### Transports

HTTP

### Your Phoenix instance; execute can run sandboxed Python and the API can change traces

 - Read and operate on projects, traces, datasets, experiments, prompts and annotations allowed by the Phoenix user
- execute runs agent-written Python inside the documented sandbox unless the code-mode flag is false

### Cost model

No separate MCP fee is stated in the opened page. Model evaluations started from Phoenix can spend provider money. None was priced.

04 / FOLLOW THE EVIDENCE

## The source trail.

Our notes are separate from the original resource. Check upstream before adopting a new version.

 - [Remote MCP documentation at reviewed revision ↗](https://github.com/Arize-ai/phoenix/blob/91ea995e0a8db1ac8b67aa2aced89ca29c722c35/docs/phoenix/integrations/remote-mcp.mdx) Checked 2026-10-14 https://github.com/Arize-ai/phoenix/blob/91ea995e0a8db1ac8b67aa2aced89ca29c722c35/docs/phoenix/integrations/remote-mcp.mdx Supports: summary, upstreamDescription, install, limitations, access, configExample, bestFor, review, compatibility, transports
- [Elastic License 2.0 ↗](https://github.com/Arize-ai/phoenix/blob/91ea995e0a8db1ac8b67aa2aced89ca29c722c35/LICENSE) Checked 2026-10-07 https://github.com/Arize-ai/phoenix/blob/91ea995e0a8db1ac8b67aa2aced89ca29c722c35/LICENSE Supports: license

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### Datadog MCP ↗ Query Datadog telemetry and operate permitted monitoring resources from an authenticated AI client.](/mcp-servers/datadog/)[### Honeycomb MCP ↗ Explore production telemetry and save investigations in Honeycomb, with additional write scope for operational changes.](/mcp-servers/honeycomb/)[### Langfuse data MCP ↗ Inspect Langfuse observations, prompts and evaluation data, and manage supported project resources through its native MCP endpoint.](/mcp-servers/langfuse/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Inspect resource evidence](/tools/)
- [Choose a host](/agents/)
- [Inspect reusable workflows](/skills/)
- [Use Undominated data](/api/)
