Multi-Agent Monitoring LangFuse MCP Server

A Model Context Protocol (MCP) server for comprehensive monitoring and observability of multi-agent systems using Langfuse.

A Model Context Protocol (MCP) server for comprehensive monitoring and observability of systems using Langfuse.

- Monitorall your agents in real-time
- Trackperformance metrics (latency, cost, token usage)
- Debugfailed executions with detailed traces
- Analyzeagent performance across time periods
- Comparedifferent agent versions via metadata filters
- Managecosts and set budget alerts
- Visualizeagent workflows

- Python 3.11 or higher
- A Langfuse account (sign up here)
- agents instrumented with Langfuse

# Install via pip pip install -r requirements.txt # Or install from source git clone https://github.com/yourusername/langfuse-mcp-python.git cd langfuse-mcp-python pip install -e .

cp .env.example .env # Edit .env and add your credentials

LANGFUSE_PUBLIC_KEY=pk-lf-xxxxx LANGFUSE_SECRET_KEY=sk-lf-xxxxx LANGFUSE_HOST=https://cloud.langfuse.com

python -m langfuse_mcp_python --transport streamable-http --host 127.0.0.1 --port 8000 --path /mcp

python -m langfuse_mcp_python --transport sse --host 127.0.0.1 --port 8000

{ "mcpServers": { "langfuse-monitor": { "command": "uvx", "args": ["--python", "3.11", "langfuse-mcp-python"], "env": { "LANGFUSE_PUBLIC_KEY": "pk-lf-xxxxx", "LANGFUSE_SECRET_KEY": "sk-lf-xxxxx", "LANGFUSE_HOST": "https://cloud.langfuse.com" } } } }

{ "mcpServers": { "langfuse-monitor": { "command": "python", "args": ["-m", "langfuse_mcp_python"], "env": { "LANGFUSE_PUBLIC_KEY": "pk-lf-xxxxx", "LANGFUSE_SECRET_KEY": "sk-lf-xxxxx" } } } }

Make sure your agents send traces to Langfuse:

from langfuse.langchain import CallbackHandler from langgraph.graph import StateGraph # Create Langfuse callback handler langfuse_handler = CallbackHandler( public_key="pk-lf-xxxxx", secret_key="sk-lf-xxxxx", host="https://cloud.langfuse.com" ) # Create your agent workflow = StateGraph(AgentState) workflow.add_node("planner", planner_node) workflow.add_node("executor", executor_node) app = workflow.compile() # Run with Langfuse monitoring result = app.invoke( {"input": "user query"}, config={ "callbacks": [langfuse_handler], "metadata": { "agent_name": "my_planner_agent", "version": "v1.0" } } )

- src/langfuse_mcp_python/server.pyCLI entrypoint and stdio transport
- src/langfuse_mcp_python/http_server.pyStreamable HTTP and SSE transport
- src/langfuse_mcp_python/utils/tool_registry.pyTool setup and registration
- src/langfuse_mcp_python/tools/Tool implementations and specs
- src/langfuse_mcp_python/integrations/langfuse_client.pyLangfuse API client
- src/langfuse_mcp_python/core/base_tool.pyShared cache and metrics

- watch_agentsMonitor active agents
- get_traceFetch a trace by ID
- analyze_performanceAggregate performance over time
- get_metricsAggregate metrics (latency, cost, tokens)

- get_scoresFetch scores
- submit_scoreCreate a score
- get_score_configsList score configurations

- get_promptsList prompts
- create_promptCreate a prompt
- delete_promptDelete a prompt

- get_datasetsList datasets
- create_datasetCreate a dataset
- create_dataset_itemAdd an item to a dataset

- get_modelsList models
- create_modelCreate a model
- delete_modelDelete a model

- get_commentsList comments
- add_commentAdd a comment

- get_annotation_queuesList annotation queues
- create_annotation_queueCreate a queue
- get_queue_itemsList queue items
- resolve_queue_itemResolve a queue item

- get_blob_storage_integrationsList integrations
- upsert_blob_storage_integrationCreate or update an integration
- get_blob_storage_integration_statusFetch integration status
- delete_blob_storage_integrationDelete an integration

- get_llm_connectionsList connections
- upsert_llm_connectionCreate or update a connection

- get_projectsList projects
- create_projectCreate a project
- update_projectUpdate a project
- delete_projectDelete a project

Show me all active agents from the last hour

Active Agent Monitoring (last_1h) Total Traces Found: 15 Showing: Top 10 traces 1. research_agent (Trace: trace-abc12...) - Status: completed - Session: session-xyz - Started: 2026-03-19T10:25:00Z - Latency: 1250ms - Tokens: 3420 - Cost: $0.0234

Watch only my research_agent and planner_agent from the last 24 hours

Analyze performance of my planner_agent over the last 24 hours

Show cost breakdown by agent for the last week

MCP Client (Claude, Cursor, etc.) -> Langfuse MCP Server (stdio/HTTP) -> Langfuse API -> Langfuse Platform -> Your Langfuse Agents

- Never commit credentials- Use environment variables
- Rotate API keysregularly
- Use read-only keyswhere possible
- Enable rate limitingin production
- Mask sensitive datain traces
- Check active agents:watch_agents
- Review performance:analyze_performance
- Check costs:get_metrics
- Investigate failures:get_trace
- Establish baseline:analyze_performancefor current version metadata
- Deploy new version with different metadata
- Compare versions by runninganalyze_performancewith version filters
- Make data-driven deployment decisions
- Track costs:get_metricsgrouped by agent
- Identify expensive agents
- Optimize high-cost operations
- Track savings over time
- Check environment variables are set correctly
- Verify Langfuse API keys are valid
- Ensure Python 3.11+ is installed
- Check logs:tail -f ~/.mcp/logs/langfuse-monitor.log
- Verify agents are instrumented with Langfuse
- Checklangfuse_handleris passed to agent invocations
- Ensure metadata includesagent_name
- Verify time window is appropriate
- Reduce number of traces fetched (use filters)
- Enable caching:CACHE_ENABLED=true
- Use "minimal" depth for trace details
- Consider batch processing for large datasets
- Fork the repository
- Create a feature branch
- Add tests for new functionality
- Submit a pull request

MIT License - see LICENSE file for details

- Langfuse- Open-source LLM observability
- LangGraph- Agent framework
- Model Context Protocol- MCP specification

Provides access to OpenTelemetry traces and metrics through Logfire.