Skip to content
Weights & Biases MCP Server

Weights & Biases MCP Server

by Weights & Biases
Released
v0.3.0
MIT
Repository Docs npm
Featured Developer Tools

Summary

The official W&B MCP server: query experiment runs, Weave LLM traces, artifacts and registries in natural language, and write findings back as a W&B report.

Features

  • Query W&B experiment runs and full run history in natural language
  • Query and count Weave LLM traces, with schema inference before querying
  • Inspect artifact versions, registries and collections, and compare versions
  • Write analyses back to W&B as shareable reports
  • Search the W&B documentation from the assistant
  • Inspect Weave agent spans and span statistics

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

  1. Open your agent's MCP configuration file.
  2. Merge the snippet below into it, filling in the environment variables with your own values.
  3. Restart the agent — the "Weights & Biases MCP Server" tools will be available.
{
  "mcpServers": {
    "wandb-mcp-server": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/wandb/wandb-mcp-server",
        "wandb_mcp_server"
      ],
      "env": {
        "WANDB_API_KEY": "<your-wandb-api-key>"
      }
    }
  }
}

Description

Weights & Biases keeps the record of what your models and LLM applications actually did — every run, sweep, artifact and trace. This is the official MCP server for that record, covering both W&B Models (classic experiment tracking) and Weave (LLM tracing and evaluation).

Rather than making you write a query language, it lets an assistant ask the questions directly: which runs in this project beat the baseline, how the loss curve moved across a sweep, which Weave traces threw errors last night, how two artifact versions differ. A schema-inference tool inspects trace structure first, so the model can build a sensible query against your data instead of guessing at field names. Around 20 tools cover run history, trace querying and counting, entity and project listing, artifact versions and comparison, model registries and collections, automations and integrations, and Weave agent spans.

The part that closes the loop is create_wandb_report_tool and log_analysis_to_wandb: an analysis an agent performs can be written back into W&B as a shareable report, so the investigation lands next to the runs it describes rather than disappearing into a chat log. A documentation search tool is also included for questions about W&B itself.

Install it with uvx from PyPI or straight from the repository, authenticate with a WANDB_API_KEY, and point it at your entity and project. Version 0.3.0, MIT licensed, maintained by Weights & Biases at wandb/wandb-mcp-server.

Related MCP Servers

MCP: Replicate MCP Server

by Replicate

New

Replicate's official MCP server: search thousands of hosted models, read their schemas, and run predictions on image, video, audio and language models from inside an agent.

Developer Tools
2 views

MCP: Hugging Face

by Hugging Face

New

Hugging Face's official remote MCP server — search and inspect models, datasets and Spaces, browse the Hub filesystem over hf:// URIs, and turn Gradio Spaces into callable tools.

Developer Tools
2 views

MCP: Resend

by Resend

Featured

Resend's official MCP server for transactional and marketing email — send, schedule and batch messages, manage templates, contacts, broadcasts, automations, domains and suppressions from an agent.

Developer Tools
7 views

MCP: Postman MCP Server

by Postman

Featured

Postman's official MCP server — reach your workspaces, collections, specs and environments from Claude Code, Cursor, VS Code or Copilot, in four tool tiers.

Developer Tools
10 views
Browse all MCP servers →