Wandb MCP Server
Weights & Biases data analysis via natural language queries in AI assistants.
Data last scanned today · Reviewed today
Overview
The Wandb MCP Server enables users to perform data analysis and debug model traces directly within their AI-powered development environment. By providing interfaces to query Weave traces and inspect experiment metrics, it allows for interactive insights into training runs and LLM performance. Users can leverage specific tools to discover trace schemas or perform complex analysis on historical data. It supports integration with various platforms like Cursor, Claude Desktop, and OpenAI, facilitating a streamlined workflow for developers who track their experiments using the Weights & Biases ecosystem.
Our verdict
This server carries a high-risk score of 40/100, primarily because it is a community-maintained project rather than an official Weights & Biases release. Because the authentication methods and potential data access patterns have not yet been editorially reviewed, caution is advised for users handling sensitive project data. Developers should evaluate it in isolated environments before integrating it into workflows containing proprietary research.
- Zero direct dependencies reduce potential supply-chain vulnerability surfaces.
- Actively maintained repository with a recent commit history.
- MIT licensed for open-source project transparency.
- Not an official product maintained by the Weights & Biases vendor.
- Authentication and OAuth security mechanisms remain unverified.
- Read-only capabilities have not undergone formal security review.
Tools
| Tool | Description | Risk |
|---|---|---|
| infer_trace_schema_tool | Discover field names, types, and sample values | low |
| query_weave_traces_tool | Analyze LLM traces with detail_level control | low |
| count_weave_traces_tool | Count traces and get storage metrics | low |
| query_wandb_tool | Query W&B runs, metrics, and experiments | low |
| get_run_history_tool | Sampled time-series metric data | low |
| create_wandb_report_tool | Create reports with markdown, charts, and panels | medium |
| log_analysis_to_wandb | Log analysis metrics to W&B as a run | medium |
| search_wandb_docs_tool | Search official W&B documentation | low |
| query_wandb_entity_projects | List projects for an entity | low |
| list_registries_tool | List model registries in an organization | low |
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| ChatGPT | ||||
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf |
Frequently asked questions
›What authentication is required to use the W&B MCP server?
The server requires a valid Weights & Biases API key, which should be provided via an Authorization header using the Bearer scheme.
›Can I restrict the server to read-only mode?
Yes, you can set the environment variable WANDB_MCP_READ_ONLY=true to disable the report creation and analysis logging tools.
›Are all tools enabled by default?
No, Weave Agents (OTel) tools are disabled by default and require setting WANDB_MCP_ENABLE_WEAVE_AGENT_TOOLS=true to enable them.
›Does the server provide access to official documentation?
Yes, it includes the search_wandb_docs_tool which allows you to query official W&B documentation for guidance.
Alternatives
Badge
Maintain this server? Add the live badge to your README.