langfuse-mcp

Overview

The langfuse-mcp server provides comprehensive access to Langfuse observability data through the Model Context Protocol. It exposes 37 tools across multiple categories including traces, observations, sessions, exceptions, prompts, datasets, annotation queues, and scores. This allows AI agents to directly query production telemetry to answer critical questions about performance failures, user sessions, prompt versions, and dataset annotations. The server is actively maintained with recent commits and offers both read-only mode and selective tool loading for efficient token usage.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• fetch_traces

  Fetch Langfuse trace data for debugging and analysis

  Try:Find traces with errors in the last 24 hoursTry:Show me slow traces with high latency
• fetch_trace

  Get details for a specific Langfuse trace

  Try:Show me details for trace abc123Try:What happened in trace 456def?
• fetch_observations

  Retrieve observations from Langfuse traces

  Try:Show all observations from yesterdayTry:Find generations with unusual outputs
• find_exceptions

  Find exceptions in Langfuse traces

  Try:Show me all exceptions in the last weekTry:Find exceptions related to authentication
• get_session_details

  Get details for a specific Langfuse session

  Try:Show me details for session xyz789Try:What happened in this user's session?
• list_prompts

  List available prompts in Langfuse

  Try:Show me all promptsTry:Find prompts labeled as production
• get_prompt

  Retrieve a specific prompt from Langfuse

  Try:Show me the latest version of my promptTry:Get prompt with ID prompt123
• list_datasets

  List available datasets in Langfuse

  Try:Show me all datasetsTry:Find datasets for testing purposes
• create_dataset_item

  Create or update a dataset item in Langfuse

  Try:Add a new item to my test datasetTry:Update existing item with ID item_123
• list_annotation_queues

  List annotation queues in Langfuse

  Try:Show me all annotation queuesTry:Find queues with pending items
• list_scores_v2

  List scores in Langfuse v2

  Try:Show me all scores for my applicationTry:Find high-scoring generations
• get_data_schema

  Get the data schema for Langfuse

  Try:Show me the Langfuse data schemaTry:What fields are available in traces?

Comparable tools

Installation

{
  "mcpServers": {
    "langfuse": {
      "command": "uvx",
      "args": ["langfuse-mcp"],
      "env": {
        "LANGFUSE_PUBLIC_KEY": "pk-...",
        "LANGFUSE_SECRET_KEY": "sk-...",
        "LANGFUSE_HOST": "https://cloud.langfuse.com"
      }
    }
  }
}

# Claude Code
code mcp add \
  -e LANGFUSE_PUBLIC_KEY=pk-... \
  -e LANGFUSE_SECRET_KEY=sk-... \
  -e LANGFUSE_HOST=https://cloud.langfuse.com \
  langfuse -- uvx langfuse-mcp

# Codex CLI
codex mcp add langfuse \
  --env LANGFUSE_PUBLIC_KEY=pk-... \
  --env LANGFUSE_SECRET_KEY=sk-... \
  --env LANGFUSE_HOST=https://cloud.langfuse.com \
  -- uvx langfuse-mcp

FAQ

How does this compare to the official Langfuse MCP server?

This server provides a full observability toolkit including traces, observations, sessions, exceptions, prompts, datasets, annotation queues, and scores. The official MCP focuses only on prompt management, making this a more comprehensive solution for debugging and analyzing agent behavior.

Can I limit the tools loaded to reduce token usage?

Yes, use selective tool loading with `langfuse-mcp --tools traces,prompts` to load only specific tool groups. Available groups include traces, observations, sessions, exceptions, prompts, datasets, annotation_queues, scores, and schema.

Compare langfuse-mcp with