The Langfuse MCP server is an implementation of the Model Context Protocol that connects your AI agent and assistants like Claude, Cursor, etc directly to your Langfuse account. It provides structured and secure access to your prompts, traces, evaluations, datasets, experiments, and metrics, so your agent can retrieve prompts, inspect observations, review scores, explore datasets, and analyze performance on your behalf.
- Prompt retrieval: Have your agent fetch text or chat prompts by name, deployment label, or version, with related prompt dependencies when needed.
- Trace and observation analysis: Instruct your agent to search generations, spans, events, agent activity, and tool calls to investigate application behavior.
- Evaluation score management: Let the agent review existing evaluation results or attach new scores to traces, sessions, dataset runs, and specific observations.
- Dataset and experiment review: Direct your agent to explore available datasets and their items, then review experiments from a chosen time range with metadata and scores.
- Metrics and evaluation resources: Have your agent query aggregate performance metrics, inspect model pricing definitions, and list annotation queues used for human evaluation.
