Use the W&B MCP Server

Connect your IDE or AI agent to the W&B Model Context Protocol (MCP) server to access your workspace data and documentation.

Model Context Protocol (MCP) lets an LLM agent query and analyze data efficiently to minimize cost in tokens. This page shows how to use the W&B MCP server to query and analyze your W&B data from your IDE or MCP client and give your client programmatic access to W&B’s documentation, so it can generate more accurate responses to W&B-related queries.

It integrates natively with most IDEs, coding clients, and chat agents, including:

• Claude
• Cursor
• Visual Studio Code (VS Code)
• Codex
• Gemini CLI
• Mistral LeChat
• Claude Desktop

The W&B MCP server supports hosted and local variants. The hosted version only supports W&B Dedicated Cloud deployments. The local version supports both Dedicated Cloud and Self-Managed deployments.

W&B MCP Server capabilities

You can use the MCP server to analyze experiments, debug traces, create reports, and get help with integrating your applications with W&B features.

The following example prompts demonstrate some of the types of tasks your agent can do when connected to the MCP server:

• Show me the top 5 runs by eval/accuracy in your-team-name/your-project-name?
• How did the latency of my hiring agent predict traces evolve over the last few months?
• Generate a wandb report comparing the decisions made by the hiring agent last month.
• How do I create a leaderboard in Weave - ask SupportBot?

Available tools

The W&B MCP server gives your agents access to the following tools:

Tool	Description	Example Query
query_wandb_tool	Query W&B runs, metrics, and experiments	”Show me runs with loss < 0.1”
query_weave_traces_tool	Analyze LLM traces and evaluations	”What’s the average latency?“
count_weave_traces_tool	Count traces and get storage metrics	”How many traces failed?“
create_wandb_report_tool	Create W&B reports programmatically	”Create a performance report”
query_wandb_entity_projects	List projects for an entity	”What projects exist?“
query_wandb_support_bot	Get help from W&B documentation	”How do I use sweeps?”

Use W&B’s remote MCP server

W&B provides a hosted MCP server at https://mcp.withwandb.com that requires no installation. The following instructions show how to configure the hosted server with various AI assistants and IDEs.

Prerequisites

• A W&B Dedicated Cloud deployment.
• A W&B API key. You can create a new one at wandb.ai/authorize.
• Set your key as an environment variable named WANDB_API_KEY.

Configure your MCP client

Select the tab containing your MCP client’s instructions:

Claude Code CLI

To add the W&B MCP server to Claude CLI, update the following command’s Authorization header with your W&B API key and run it in your terminal:

claude mcp add --transport http wandb https://mcp.withwandb.com/mcp \
  --header "Authorization: Bearer <your-wandb-api-key>"

Add --scope user for a global configuration, or omit it to configure for the current project only.

For more detailed information, see Claude’s documentation.

Claude Code Desktop

Claude Desktop’s built-in custom connectors interface does not support API key authentication for remote MCP servers. To work around this, use the mcp-remote npm proxy to connect Claude Desktop to the W&B remote MCP server. The proxy runs locally and forwards requests to https://mcp.withwandb.com/mcp with your Bearer token.

You need Node.js installed on your system.

Open your Claude Desktop config file in a text editor. You can find the config file at the following location for your OS:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the following to the JSON object in your config file, replacing <your-wandb-api-key> with your W&B API key:

{
  "mcpServers": {
    "wandb": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.withwandb.com/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer <your-wandb-api-key>"
      }
    }
  }
}

The full header value is set through the env field rather than directly in args to work around a space-escaping issue in some Claude Desktop versions.

Restart Claude Desktop to activate the new configuration. Verify the connection by entering the prompt “List the projects in my W&B account.”

Codex

To add the W&B MCP server to Codex, update the following command’s --bearer-token-env-var argument with the environment variable containing your W&B API key, then run it in your terminal:

export WANDB_API_KEY=<your-wandb-api-key>
codex mcp add wandb --url https://mcp.withwandb.com/mcp --bearer-token-env-var <your-wandb-api-key-environment-variable>

Cursor

You can install the W&B server in Cursor automatically using a one-click installation link (requires adding Bearer <your-wandb-api-key> in the Authorization field), or manually using the following instructions:

1. On macOS, open the Cursor menu, select Settings, and then select Cursor Settings. On Windows or Linux, open the Preferences menu, select Settings, and then select Cursor Settings.
2. From the Cursor Settings menu, select Tools and MCP. This opens the Tools menu.
3. In the Installed MCP Servers section, select Add Custom MCP. This opens the mcp.json configuration file.
4. In the configuration file, in the mcpServers JSON object, add the following wandb object:

{
  "mcpServers": {
    "wandb": {
      "transport": "http",
      "url": "https://mcp.withwandb.com/mcp",
      "headers": {
        "Authorization": "Bearer <your-wandb-api-key>",
        "Accept": "application/json, text/event-stream"
      }
    }
  }
}

5. Restart Cursor to make the changes take effect.
6. Verify that the chat agent has access to the W&B MCP server by entering the prompt “List the projects in my W&B account.”

For more detailed information, see Cursor’s documentation.

Gemini CLI

To add the W&B MCP server to Gemini CLI:

1. Install the W&B MCP extension with a single command:

   # Install the extension
   gemini extensions install https://github.com/wandb/wandb-mcp-server
   

2. Once installed, restart the Gemini CLI.

3. Verify that the chat agent has access to the W&B MCP server by entering the prompt “List the projects in my W&B account.”

For more detailed information, see Gemini’s documentation.

Mistral LeChat

To add the W&B MCP server to Mistral LeChat:

1. From the Intelligence menu, select Add Connector to open the Connector window.

2. Select the Custom MCP Connector tab.

3. Configure the fields using the following values:

   • Connector Server: https://mcp.withwandb.com/mcp
   • Description: (Optional) A brief arbitrary description of the connection.
   • Authentication Method: Select API Token Authentication. This opens additional fields.
   • Header name: Leave the default value, Authorization.
   • Header type: Select Bearer.
   • Header value: Enter your W&B API token.

4. Once you have configured all the fields, select Create. LeChat adds the MCP server to your configuration.

5. Verify that the chat agent has access to the W&B MCP server by entering the prompt “List the projects in my W&B account.”

For more detailed information, see LeChat’s documentation.

OpenAI

To add the W&B MCP server to your OpenAI calls, add the server’s information to the tools field of your OpenAI responses configuration:

from openai import OpenAI
import os

client = OpenAI()

resp = client.responses.create(
    model="gpt-4o",
    tools=[{
        "type": "mcp",
        "server_label": "wandb",
        "server_description": "Query W&B data",
        "server_url": "https://mcp.withwandb.com/mcp",
        "authorization": os.getenv("<your-wandb-api-key>"),
        "require_approval": "never",
    }],
    input="List the projects in my W&B account.",
)

print(resp.output_text)

Set up a local version of the W&B MCP server

If you need to run the MCP server locally for W&B Self-Managed deployments, development, testing, or air-gapped environments, you can install and run it on your machine.

Prerequisites

• A W&B API key. You can create a new one at wandb.ai/authorize.
• Set your key as an environment variable named WANDB_API_KEY.
• Set the WANDB_BASE_URL environment variable if you are using W&B Self-Managed.
• Python 3.10 or higher
• uv (recommended) or pip

See uv’s docs for installation instructions.

Install and configure the MCP server

To install the MPC server locally:

To install the W&B MCP server on your local machine, use one of the following installation commands:

uv

uv pip install wandb-mcp-server

pip

pip install wandb-mcp-server

Install directly from GitHub

pip install git+https://github.com/wandb/wandb-mcp-server

Once you have installed the MCP server locally, configure your MCP client to use it. Select an MCP client to continue:

Claude Code CLI

Run the following command in your terminal. Add --scope user for a global configuration, or omit it to configure for the current project only.

claude mcp add wandb \
  -e WANDB_API_KEY=your-api-key \
  -e WANDB_BASE_URL=https://your-wandb-instance.example.com \
  -- uvx --from git+https://github.com/wandb/wandb-mcp-server wandb_mcp_server

Claude Code Desktop

Open your Claude Code Desktop config file in a text editor. You can find the config file at the locations for your OS:

• macOS: ~/Library/Application\ Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the following to your JSON object to your Claude Code Desktop config file. Use the full path to uvx because Claude Code Desktop may not find your uvx installation otherwise.

{
  "mcpServers": {
    "wandb": {
      "command": "/full/path/to/uvx",
      "args": [
        "--from",
        "git+https://github.com/wandb/wandb-mcp-server",
        "wandb_mcp_server"
      ],
      "env": {
        "WANDB_API_KEY": "<your-wandb-api-key>",
        "WANDB_BASE_URL": "https://your-wandb-instance.example.com"
      }
    }
  }
}

Restart Claude Desktop to activate the new configuration.

Codex

Run the following command in your terminal:

codex mcp add wandb \
  --env WANDB_API_KEY=your_api_key_here \
  --env WANDB_BASE_URL=https://your-wandb-instance.example.com \
  -- uvx --from git+https://github.com/wandb/wandb-mcp-server wandb_mcp_server

Cursor

Add the following to your mcp.json configuration:

{
  "mcpServers": {
    "wandb": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/wandb/wandb-mcp-server",
        "wandb_mcp_server"
      ],
      "env": {
        "WANDB_API_KEY": "<your-wandb-api-key>",
        "WANDB_BASE_URL": "https://your-wandb-instance.example.com"
      }
    }
  }
}

VS Code

Add the following to your .vscode/mcp.json or global MCP configuration:

{
  "servers": {
    "wandb": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/wandb/wandb-mcp-server",
        "wandb_mcp_server"
      ],
      "env": {
        "WANDB_API_KEY": "<your-wandb-api-key>",
        "WANDB_BASE_URL": "https://your-wandb-instance.example.com"
      }
    }
  }
}

For web-based clients or testing, run the server with HTTP transport:

uvx wandb_mcp_server --transport http --host 0.0.0.0 --port 8080

To expose the local server to external clients like OpenAI, use ngrok:

uvx wandb_mcp_server --transport http --port 8080

# In another terminal, expose with ngrok
ngrok http 8080

If you expose the server using ngrok, update your MCP client configuration to use the ngrok URL.

Usage tips

• Provide your W&B project and entity name: Specify the W&B entity and project in your queries for accurate results.
• Avoid overly broad questions: Instead of “what is my best evaluation?”, ask “what eval had the highest f1 score?”
• Verify data retrieval: When asking broad questions like “what are my best performing runs?”, ask the assistant to confirm it retrieved all available runs.