W&B MCP Server
Queries Weights & Biases runs and Weave traces in plain language.
Add it to your client
Officialuvx wandb-mcp-server
{
"mcpServers": {
"wandb-mcp": {
"command": "uvx",
"args": [
"wandb-mcp-server"
]
}
}
}- Transport
- both
- Runtime
- Python
- Auth
- Required
- License
- MIT
W&B MCP Server connects MCP clients to Weights & Biases, letting agents query run metrics, sampling histories, and LLM traces using natural language. The server reads from your W&B workspace and can write analysis back as shareable reports. It runs either as a hosted HTTP endpoint or locally via stdio with uvx.
Highlights
- Query run metrics and sampling histories in natural language
- Inspect and count LLM traces from Weave
- Infer trace field schemas automatically
- Write analysis results back to W&B as shareable reports
- Hosted HTTP or local stdio deployment options
Ideal for ML teams using Weights & Biases to explore experiments and traces programmatically. Requires a W&B API key; both deployment modes are free to use.
Similar MCP servers
AWS MCP Servers
MCP serverA large collection of AWS-specific MCP servers, installed one at a time.
LangSmith MCP Server
MCP serverOpens LangSmith traces, prompts, datasets and experiments to an agent.
Wolfram Alpha MCP Server
MCP serverQueries the Wolfram Alpha API for computation and curated facts, returning structured answers rather than prose.

Comments(0)
Sign in to comment