Langfuse Docs MCP Server for Claude
Connect Claude Code, Cursor, Copilot, Windsurf, and other MCP clients to the public Langfuse documentation MCP server for tracing, prompt management, evaluation, and agent observability implementation help.
Open the source and read safety notes before installing.
Safety notes
- The public Docs MCP endpoint is unauthenticated and documentation-only; it should not be treated as a connection to your Langfuse project data.
- Langfuse also documents an authenticated project MCP server at /api/public/mcp; that separate endpoint exposes read and write tools by default, so restrict write tools if you need read-only access.
- MCP tool schemas are self-describing and may evolve; clients should inspect live capabilities instead of hard-coding a fixed tool list.
- Use human review for generated tracing, prompt, evaluation, or production-debugging changes before shipping them.
Privacy notes
- Queries sent to the public Docs MCP endpoint and search-docs API are processed by Langfuse's hosted documentation service.
- Avoid including proprietary code, customer data, prompts, traces, or API keys in documentation-search queries.
- The authenticated Langfuse platform MCP uses Basic Auth with project-scoped public and secret keys; never paste the raw key pair or base64 header into prompts, commits, screenshots, or shared logs.
- If you connect the platform MCP, observations, prompt content, datasets, scores, comments, metrics, model metadata, and media from that project can be surfaced to the MCP client and model provider.
Prerequisites
- Claude Code, Cursor, GitHub Copilot in VS Code, Windsurf, or another MCP client with Streamable HTTP support.
- Network access to https://langfuse.com/api/mcp.
- For clients without Streamable HTTP support, Node.js and npx if you proxy the endpoint through mcp-remote.
- A separate Langfuse project and API keys only if you also connect the authenticated Langfuse platform MCP server.
Schema details
- Install type
- cli
- Troubleshooting
- No
- Estimated setup
- 3 minutes
- Difficulty
- beginner
Full copyable content
claude mcp add --transport http langfuse-docs https://langfuse.com/api/mcp --scope userAbout this resource
Overview
Langfuse publishes a public Docs MCP server at https://langfuse.com/api/mcp.
It exposes Langfuse documentation search and page retrieval to MCP clients, so
Claude Code, Cursor, Copilot, Windsurf, and other agents can find current
Langfuse implementation guidance without scraping the website by hand.
Use this listing for the unauthenticated docs endpoint. Langfuse also documents
an authenticated platform MCP server for project data at /api/public/mcp; that
server is useful for prompts, traces, datasets, metrics, comments, evaluators,
and scores, but it carries real project authority and should be configured more
carefully.
Server Types
| Server | Endpoint | Authentication | Primary Use |
|---|---|---|---|
| Langfuse Docs MCP | https://langfuse.com/api/mcp |
None | Search Langfuse docs, fetch pages, retrieve a docs overview |
| Langfuse Cloud MCP | https://cloud.langfuse.com/api/public/mcp |
Basic Auth with project API keys | Work with project prompts, observations, datasets, scores, metrics, comments, models, media, evaluators, and health |
| Regional Langfuse Cloud MCP | US, Japan, and HIPAA cloud endpoints | Basic Auth with project API keys | Same platform tools in a regional cloud environment |
| Self-hosted Langfuse MCP | https://your-domain.com/api/public/mcp |
Basic Auth with project API keys | Platform MCP for your self-hosted deployment |
Install In Claude Code
Add the public docs server:
claude mcp add \
--transport http \
langfuse-docs \
https://langfuse.com/api/mcp \
--scope user
Then start Claude Code and run /mcp to confirm the langfuse-docs server is
connected.
Manual Configuration
Add this to a Claude settings file:
{
"mcpServers": {
"langfuse-docs": {
"transportType": "http",
"url": "https://langfuse.com/api/mcp",
"verifySsl": true
}
}
}
For clients that expect stdio rather than Streamable HTTP, Langfuse documents
mcp-remote as a local proxy pattern:
{
"mcpServers": {
"langfuse-docs": {
"command": "npx",
"args": ["mcp-remote", "https://langfuse.com/api/mcp"]
}
}
}
Docs Tools
The MCP reference currently identifies three public docs tools:
| Tool | Purpose |
|---|---|
searchLangfuseDocs |
Search Langfuse documentation, GitHub issues, and discussions for a query |
getLangfuseDocsPage |
Fetch a specific documentation page |
getLangfuseOverview |
Retrieve a concise overview of Langfuse documentation |
The same docs search is also exposed as a REST endpoint at
https://langfuse.com/api/search-docs for lightweight non-MCP workflows.
Authenticated Platform MCP
When an agent needs project data instead of docs, configure the authenticated Langfuse MCP server with a project-scoped public key and secret key encoded in a Basic Auth header:
claude mcp add --transport http langfuse https://cloud.langfuse.com/api/public/mcp \
--header "Authorization: Basic <base64(LANGFUSE_PUBLIC_KEY:LANGFUSE_SECRET_KEY)>"
Langfuse documents regional cloud endpoints for US, Japan, and HIPAA US regions,
plus a self-hosted endpoint at https://your-domain.com/api/public/mcp.
The platform MCP is broader than the docs endpoint. The MCP reference lists tools for prompt lookup and creation, prompt-label updates, observations, annotation queues, comments, datasets, dataset items, dataset runs, scores, score configs, metrics, models, media, evaluators, evaluation rules, and health checks. Both read and write tools are available by default, so configure your client's tool allowlist if the agent should only inspect data.
Use Cases
- Ask Claude Code to find the current Langfuse tracing setup for a framework.
- Let Cursor retrieve the right docs page before instrumenting an existing OpenAI, LangChain, LangGraph, or agent workflow.
- Compare prompt-management, evaluation, and observability docs without leaving the editor.
- Use the authenticated platform MCP to list prompts, inspect observations, pull dataset items, or create evaluation artifacts after project credentials are scoped and reviewed.
- Pair the Docs MCP with the Langfuse Agent Skill when an agent can benefit from both source-aware documentation search and a repeatable implementation playbook.
Safety and Privacy
Keep the two server surfaces separate. The public Docs MCP is useful context for implementation and troubleshooting, but it does not prove anything about your own Langfuse project. The authenticated platform MCP can access real project data and may create or update resources such as prompts, scores, comments, datasets, evaluators, or evaluation rules.
For production projects, prefer project-scoped keys, store the Basic Auth header in local MCP configuration instead of prompts, and restrict write tools when the agent only needs investigation access. Review generated code before enabling tracing in production because traces can capture prompts, model outputs, tool arguments, costs, latency, metadata, and customer-sensitive payloads.
Source Review
Verified on 2026-06-18:
- Langfuse's Docs MCP page documents
https://langfuse.com/api/mcp, Streamable HTTP transport, no authentication, Claude Code setup, Cursor, Copilot, Windsurf,mcp-remote, MCP reference links, and the RESTsearch-docsendpoint. - Langfuse's API and data-platform MCP page documents the authenticated
/api/public/mcpendpoint, regional cloud endpoints, self-hosted endpoint, Basic Auth setup, and the warning that read and write tools are available by default unless the client restricts them. - The Langfuse MCP reference documents a public Docs MCP server and an authenticated Cloud MCP server, including docs tools and platform tool categories.
- The existing HeyClaude Langfuse entry is a general tools listing for the observability platform. This entry is limited to the MCP connection surfaces, with the public docs endpoint as its canonical source.
Duplicate Check
Checked current content/mcp/, content/tools/, content/skills/, open pull
requests, and prior closed PR history for Langfuse MCP, langfuse-docs,
https://langfuse.com/api/mcp, https://cloud.langfuse.com/api/public/mcp, and
mcp.reference.langfuse.com. The catalog already has content/tools/langfuse.mdx
for the general Langfuse platform. A prior closed MCP attempt used the generic
Langfuse repository as its canonical source and was rejected as same-source
ambiguous. This resubmission is narrowed to the official Docs MCP and MCP
reference sources, with explicit separation from the general tools listing.
Source citations
Add this badge to your README
How it compares
Langfuse Docs MCP Server for Claude side by side with 3 alternatives on trust, install, platform support, and disclosed safety notes — all from reviewed registry metadata.
| Field | Langfuse Docs MCP Server for Claude Connect Claude Code, Cursor, Copilot, Windsurf, and other MCP clients to the public Langfuse documentation MCP server for tracing, prompt management, evaluation, and agent observability implementation help. Open dossier | Arize Phoenix MCP Server for Claude Inspect LLM traces and spans, manage prompts, explore datasets, and review evaluation experiments from Claude — with the official Arize Phoenix MCP server, built into the open-source Phoenix AI observability platform. Open dossier | Opik MCP Server for Claude Debug, evaluate, and monitor LLM applications from Claude — read traces and spans, score outputs, save prompts, run evaluation experiments, and query project metrics — with the official Opik MCP server by Comet. Open dossier | Honeycomb MCP Server for Claude Connect Claude to Honeycomb observability data — query traces and events, investigate alerts, manage boards and triggers, create SLOs, and cross-reference production behavior with your codebase — with the official Honeycomb hosted MCP server. Open dossier |
|---|---|---|---|---|
| Trust | ||||
| Install risk | Review first | Review first | Review first | Review first |
| Notes | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ |
| Category | mcp | mcp | mcp | mcp |
| Source | source-backed | source-backed | source-backed | source-backed |
| Author | Langfuse | Arize AI | Comet | Honeycomb |
| Added | 2026-06-18 | 2026-06-18 | 2026-06-18 | 2026-06-18 |
| Platforms | Claude CodeCodexCursorClaude Desktop | Claude CodeClaude Desktop | Claude CodeClaude Desktop | Claude CodeClaude Desktop |
| Source repo | — | — | — | — |
| Safety notes | ✓The public Docs MCP endpoint is unauthenticated and documentation-only; it should not be treated as a connection to your Langfuse project data. Langfuse also documents an authenticated project MCP server at /api/public/mcp; that separate endpoint exposes read and write tools by default, so restrict write tools if you need read-only access. MCP tool schemas are self-describing and may evolve; clients should inspect live capabilities instead of hard-coding a fixed tool list. Use human review for generated tracing, prompt, evaluation, or production-debugging changes before shipping them. | ✓The `write` and upsert tools can create or modify prompts and add dataset examples — review Claude's proposed changes before confirming. For self-hosted deployments, ensure your Phoenix instance is accessible from the machine running the MCP server. | ✓The `write` tool can create traces, spans, scores, and prompts in your Opik workspace — review any write operations before confirming. The `run_experiment` tool executes evaluation experiments end-to-end, which may incur LLM API costs depending on your experiment configuration. | ✓The server is hosted by Honeycomb (AWS-backed) and authenticated via OAuth 2.1 or API key — write-scoped access can create/update Boards, Triggers, and SLOs. Writing to Honeycomb resources (boards, triggers, SLOs) requires the `create` scope; verify you grant only the scopes your workflow needs. Canvas investigations and alert management are write operations — review Claude's proposed changes before executing in production environments. |
| Privacy notes | ✓Queries sent to the public Docs MCP endpoint and search-docs API are processed by Langfuse's hosted documentation service. Avoid including proprietary code, customer data, prompts, traces, or API keys in documentation-search queries. The authenticated Langfuse platform MCP uses Basic Auth with project-scoped public and secret keys; never paste the raw key pair or base64 header into prompts, commits, screenshots, or shared logs. If you connect the platform MCP, observations, prompt content, datasets, scores, comments, metrics, model metadata, and media from that project can be surfaced to the MCP client and model provider. | ✓LLM traces (including prompts and completions), evaluation annotations, session data, datasets, and experiment results from your Phoenix workspace are surfaced in Claude's context. Your `PHOENIX_API_KEY` is passed as an environment variable — store it in a secrets manager rather than plaintext shell config. | ✓LLM traces (including prompts and completions), evaluation scores, experiment results, and prompt library contents from your Opik workspace are surfaced in Claude's context. Your `OPIK_API_KEY` is passed as an environment variable — store it in a secrets manager rather than plaintext shell config. | ✓Trace data, event fields, alert details, and query results from Honeycomb are sent through Honeycomb's hosted MCP endpoint and surfaced in Claude's context. Honeycomb API keys (`KEY_ID:SECRET_KEY` format) are secrets — store them in your MCP client config or environment, never in repositories. |
| Prerequisites |
|
|
|
|
| Install | | | | |
| Config | | | | |
| Citations | ||||
| Claim | Unclaimed | Unclaimed | Unclaimed | Unclaimed |
Signals
Loading live community signals…
A short, calm digest of reviewed Claude resources. Unsubscribe any time.