Advanced12 min
LangSmith MCP Server
The LangSmith MCP Server exposes your entire observability workspace as callable tools via Model Context Protocol — query traces with FQL, manage datasets, push prompts, and check billing from Claude Desktop, Cursor, or any custom agent without leaving your editor.
Quick Reference
- →Hosted server at https://langsmith-mcp-server.onrender.com/mcp — add your API key and connect immediately
- →Auth header is LANGSMITH-API-KEY (not X-API-Key) — pass your lsv2_pt_... key
- →Primary trace tool: fetch_runs with FQL filtering — use eq(), gt(), has(), and() operators
- →Page-number pagination — default 25,000 chars/page, max 30,000; pass page=2 to get the next page
- →Dataset tools: list_datasets, read_dataset, list_examples, read_example, create_dataset, update_examples
- →Prompt tools: list_prompts, get_prompt_by_name, push_prompt | Thread: get_thread_history | Billing: get_billing_usage
- →Self-hosted Docker: use LANGSMITH_ENDPOINT env var (not LANGSMITH_API_URL), server runs on port 8000
Should You Use the LangSmith MCP Server?
The LangSmith MCP Server is the right tool when you want to query your observability workspace conversationally or let an agent introspect its own production history. It is not the right tool for every LangSmith task.
| Access Path | When to Use | What It Gives You |
|---|---|---|
| LangSmith UI | Visual exploration, dashboards, annotation queues | Full-featured browser UI — charts, filters, side-by-side evals |
| LangSmith SDK | Programmatic pipelines, CI eval harness, custom scripts | Full API coverage, strongly typed, best for automation |
| MCP Server | Conversational queries from an AI client, agent self-monitoring | Natural-language access to traces/datasets/prompts from any MCP client |
| LangSmith Fetch CLI | Terminal-native one-liners, quick trace dumps without an MCP client | Lightweight CLI — no AI client needed, just your terminal |
Best use case
The MCP server shines when you want to ask Claude Desktop or Cursor 'what was the error rate for my support agent last hour' and get an answer without switching apps. For batch evaluation, use the SDK. For quick terminal lookups, use the Fetch CLI.