Built-in MCP Server with OAuth

Overview

LangWatch now ships a built-in MCP (Model Context Protocol) server, mounted directly in the app with OAuth PKCE authorization. Connect your IDE or AI coding assistant - Claude Code, Cursor, or any MCP-compatible client - and manage your LangWatch project without leaving your editor. Query analytics, manage datasets and evaluators, tag prompts, and more.

What You Can Do Now

• Connect from any MCP client - Claude Code, Cursor, and other MCP-compatible tools can authenticate via OAuth PKCE and interact with your LangWatch project.

• Manage datasets - create, read, update, and delete datasets and records directly from your editor.

• Manage evaluators - list, create, and configure evaluators without opening the dashboard.

• Query analytics - pull metrics, search traces, and inspect individual traces from your terminal.

• Tag prompts - assign and manage prompt tags via MCP tools, keeping your prompt versioning workflow in your editor.

• 30-day session persistence - sessions are backed by Redis for multi-pod support, so you don't have to re-authenticate constantly.

Rollout

Available now for all LangWatch users. See the MCP Server documentation for setup instructions.

Notes

• Sessions persist for 30 days via Redis-backed storage, with proper 401 responses on expiration.

• The MCP server supports multi-pod deployments out of the box.

• Dataset CRUD and evaluator management are available in both the TypeScript and Python SDKs as well.