LangSmith

Connects Claude directly to LangSmith's observability platform through a full set of read and write operations. You get tools for pulling conversation histories from threads, fetching traces and runs with FQL filtering, managing prompts and datasets, listing experiments with metrics, and checking billing usage. Everything uses character-based pagination to keep responses within token budgets. Useful when you're debugging production LLM apps, analyzing eval results across experiments, or building agents that need to reference past conversations and prompt templates. Ships with a hosted version over HTTP, or run it locally via uv with your LangSmith API key. Built by the LangChain team for both cloud and self-hosted LangSmith instances.