Awesome MCP › Software Development

avivsinai/langfuse-mcp

⭐ 105 Python repository created 2025-03-21

Langfuse MCP Server gives MCP clients read access to Langfuse, the open-source observability platform for language-model applications. Langfuse records traces of model calls — prompts, completions, tool invocations, token counts, latencies, costs and errors — and this server makes that history queryable as tools: finding traces by criteria, drilling into a specific trace's observations, examining sessions that group related traces, and retrieving error and exception detail. The characteristic use is self-debugging: an agent whose earlier run behaved wrongly can look up the actual trace and see which tool call returned unexpected output or where token use spiked, rather than the developer reading through a web dashboard and pasting findings back into the conversation. It runs locally against a Langfuse instance, cloud or self-hosted, with project credentials. The project is Python, supports versions 3.10 through 3.14, is published on PyPI as langfuse-mcp with substantial monthly download figures, and is MIT licensed. It ships agent skills for Claude Code and Codex alongside the plain MCP server. The audience is teams running language-model applications instrumented with Langfuse who want failure investigation to happen inside the development session where the fix will be written.

https://github.com/avivsinai/langfuse-mcp

mcplangfuseobservabilitytracingpythonllm-debuggingdeveloper-tools