---
title: "Langfuse data MCP: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/mcp-servers/langfuse/
description: "Inspect Langfuse observations, prompts and evaluation data, and manage supported project resources through its native MCP endpoint."
---

# Langfuse data MCP: review, setup & limitations · Undominated.ai

> Inspect Langfuse observations, prompts and evaluation data, and manage supported project resources through its native MCP endpoint.

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

OBSERVABILITY / Langfuse

# Langfuse data MCP

Inspect Langfuse observations, prompts and evaluation data, and manage supported project resources through its native MCP endpoint.

 [See setup guidance ↓](#setup)[Original source ↗](https://langfuse.com/docs/api-and-data-platform/features/mcp-server)

SOURCE REVIEW

 Reviewed 2026-10-14
 Evidence 2 linked sources
 Publisher Langfuse
 Licence Not established
 [Read what was—and wasn’t—checked ↓](#review)

“Both read and write tools are available by default.”

 [Langfuse · upstream description ↗](https://langfuse.com/docs/api-and-data-platform/features/mcp-server) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - Brings prompts, observations, scores and evaluation datasets into an agent’s investigation.
- Project-scoped API credentials and regional endpoints make the data boundary explicit.
- Langfuse documents a stateless project-scoped MCP server that uses an API key, with both read and write tools available unless the client allowlists read-only tools.
- The reference site separates that authenticated project server from an unauthenticated documentation server and tells clients not to assume a fixed tool list.

### A good fit for

 - Investigate an LLM application trace or prompt.
- Manage reviewed evaluation datasets, scores and annotation work.
- Managing prompts in one Langfuse Cloud project with a project API key
- Searching public Langfuse documentation without a project credential

### Weigh up before choosing

 - Read and write tools are available by default; use a client allowlist to restrict mutations.
- This authenticated data server is different from Langfuse’s public documentation MCP.
- Self-hosted deployments behind a reverse proxy or internal hostname may need to preserve the public Host header or configure LANGFUSE_MCP_ALLOWED_HOSTS; cloud endpoints differ by region.
- The setup page says Pi does not ship with built-in MCP support and points at a community adapter.
- Write tools are available by default; read-only use depends on the client allowlist, which this server page does not say it enforces itself.
- No licence was stated on these pages, and none was inferred.
- This review did not execute the resource.
- The server was not called.
- Client allowlisting is a host-side restriction, not a read-only API key; do not confuse logical-root filtering with parent-id filtering.

02 / THE REVIEW RECORD

## What we actually inspected.

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

### Material inspected

 - Native Langfuse MCP setup guide and generated MCP Reference.
- https://langfuse.com/docs/api-and-data-platform/features/mcp-server
- https://mcp.reference.langfuse.com
- Round 2 reread of acquired publisher setup and permission evidence; see immutable wave review.

### Our findings

 - The server uses stateless Streamable HTTP with project API-key Basic authentication.
- The current reference includes prompt, dataset, score and annotation mutations; it is broader than prompt retrieval.
- The Langfuse data MCP server is stateless, and each API key is scoped to one project. The reference calls that endpoint an authenticated project-scoped server.
- Read and write tools are available by default. Restricting writes is described as a client allowlist, not as a separate server mode on this page.
- A second, unauthenticated server is documented for documentation search. Clients are told to inspect tools dynamically. The page says Pi does not include built-in MCP support.
- The current project MCP exposes reads and writes by default; observation tools distinguish logical roots from physical parent relationships.

### Not established by this review

 - Server launch, authenticated tool calls, and release-to-source parity were not tested.
- The resource was not executed and its tools were not called.
- The server was not called. A source review is not a runtime or security certification.

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://langfuse.com/docs/api-and-data-platform/features/mcp-server)
 - Choose the documented cloud-region endpoint, or your self-hosted /api/public/mcp endpoint.
- Create project API keys and configure Basic authentication using the official client instructions; keep secrets outside shared project files.
- Apply a read-only tool allowlist if the agent should investigate without changing project data.

### Before you start

 - Langfuse project API keys.
- A client supporting Streamable HTTP and the required authorization header.

### Documented configuration example

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

 {
 "mcpServers": {
 "langfuse-cloud": {
 "url": "https://cloud.langfuse.com/api/public/mcp",
 "headers": {
 "Authorization": "Basic <base64(LANGFUSE_PUBLIC_KEY:LANGFUSE_SECRET_KEY)>"
 }
 }
 }
} Copy configuration ↗

### Compatibility

Claude Code · Cursor · Codex · Streamable HTTP clients

### Transports

Streamable HTTP

### Project API-key Basic authentication

 - Returns project prompts, observations, evaluation records and related data.
- Can change supported prompt, dataset, scoring and annotation resources.

### Cost model

Langfuse cloud plan or self-hosting costs apply; the connected model client is separate.

04 / FOLLOW THE EVIDENCE

## The source trail.

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

 - [Langfuse data MCP setup and scope ↗](https://langfuse.com/docs/api-and-data-platform/features/mcp-server) Checked 2026-10-14 https://langfuse.com/docs/api-and-data-platform/features/mcp-server Supports: summary, upstreamDescription, whySelected, bestFor, limitations, install, access, compatibility, transports, review
- [Langfuse generated MCP tool reference ↗](https://mcp.reference.langfuse.com) Checked 2026-10-14 https://mcp.reference.langfuse.com Supports: summary, whySelected, bestFor, access, limitations, review

KEEP COMPARING

## Other approaches to consider.

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

 [### Apache SkyWalking MCP Server ↗ Queries a SkyWalking OAP you already run. The inspected install builds a local image; it does not pull a published one.](/mcp-servers/apache-skywalking-mcp/)[### AWS Managed Prometheus MCP ↗ Pins the AWS Labs server for PromQL against Amazon Managed Prometheus, and records that the basic Kiro example omits the workspace URL.](/mcp-servers/awslabs-prometheus/)[### Axiom MCP ↗ Queries Axiom telemetry and manages observability resources through an account-authorized hosted connection.](/mcp-servers/axiom/)

 [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/)
