Weights & Biases MCP Server
@wandb
The official implementation of the W&B Models and Weave MCP server.
Overview
What is Weights & Biases MCP Server?
Query and analyze your Weights & Biases data using natural language through the Model Context Protocol. The server provides 16 tools to interact with W&B experiments, Weave traces, reports, artifacts, model registries, documentation, and automations. It is designed for ML practitioners who want to explore and debug their W&B data from AI assistants and IDEs.
How to use Weights & Biases MCP Server?
Use the hosted server at https://mcp.withwandb.com with a W&B API key from wandb.ai/authorize. One-click or one-command setup is available for Cursor, Claude Code, OpenAI Codex, and Gemini CLI. Manual configuration is provided for Claude Desktop, VSCode, OpenAI Responses API, and Mistral Chat. For local development, run uvx with the git repository via STDIO.
Key features of Weights & Biases MCP Server
- Query W&B runs, metrics, and experiments via natural language
- Analyze LLM traces with configurable detail levels
- Create reports with charts, markdown, and custom panels
- Search and explore Weave trace schemas
- Compare and inspect artifact versions
- List model registries, collections, and automations
- Proxy W&B documentation search through a single connection
Use cases of Weights & Biases MCP Server
- Analyze top runs by evaluation metrics from a specific project
- Debug trace latency trends across months of production data
- Generate a W&B report comparing model decisions over time
- Explore artifact version histories and diff between model versions
- Query W&B documentation for how-to guidance without leaving the chat
FAQ from Weights & Biases MCP Server
How do I get started?
Get your API key from wandb.ai/authorize and use the hosted server at https://mcp.withwandb.com with one-click installation in Cursor or a one-command setup in Claude Code, OpenAI Codex, or Gemini CLI.
What authentication is required?
The server uses Bearer token authentication with your W
