Langfuse

Langfuse develops an open-source platform for LLM observability, prompt management, evaluation, and analytics, and operates the authenticated Langfuse MCP Server for project-scoped data platform access. The goal is to make AI applications traceable throughout the development lifecycle, from prototype to production.

Hierarchical tracing captures every LLM call, tool invocation, and retrieval step in an application, making it possible to pinpoint failure sources in complex agent chains. To evaluate output quality, the platform offers LLM-as-a-judge methods, heuristic functions, and human review. Prompt management separates prompts from code and allows one-click deployment of new versions, while experiments and datasets support systematic side-by-side comparison of test cases. Cost and latency monitoring with dashboards and alerts, plus collaborative annotation workflows for building golden datasets, round out the platform.

The Langfuse MCP Server gives IDE agents structured access to manage prompts, query traces, and use Langfuse data directly from the development environment.

The business model is hybrid: the entire codebase is MIT-licensed and free to self-host, while Langfuse Cloud offers a managed service with usage-based tiered pricing and a credit-card-free hobby tier. The company reports more than 50,000 companies using the platform, processing over 90 billion observations per month combined.

Langfuse was founded by Marc Klingen, Maximilian Deichmann, and Clemens Rawert, headquartered in Berlin with a smaller office in San Francisco. The founders identified the typical pain points of debugging LLM applications during Y Combinator's Winter 2023 batch; in January 2026, Langfuse was acquired by ClickHouse to accelerate its development further.