• MONITOR
  • AUDIT
  • COMPARE
  • HUMAN

How to track experiment metrics with your AI

Connect Weights & Biases to your agent to monitor training progress without leaving your chat interface.

Ask AI about this page

Short answer

How can I track my experiment metrics using AI?

Your agent uses Weights & Biases to fetch specific run data, including loss curves and hyperparameter configurations. You can ask for summaries of recent runs or deep dives into specific training sessions. This gives you a direct line into your experiment history.

The outcomes.

What you get from your agent.

Your agent handles the data retrieval so you can focus on the results.

MONITOR

Track live progress

The AI pulls current loss and accuracy values from active runs. You get immediate updates on how your model is performing.

AUDIT

Check configurations

Ask for the specific hyperparameters used in any past run. Your agent finds the exact settings that led to a specific result.

COMPARE

Analyze run history

Compare different training sessions to see which configurations worked best. Your agent summarizes the differences in performance.

HUMAN

Decide next steps

The AI provides the data, but you decide if a run is worth saving or if you need to adjust your learning rate.

The workflow

What your AI does when you ask for data.

The agent acts as a bridge between your chat and your experiment dashboard.

  1. Find your runs

    The agent searches through your existing projects to locate the specific training sessions you want to discuss.

    list_project_runs

  2. Fetch run data

    Once the right run is identified, the agent pulls the detailed metrics and metadata for that specific ID.

    get_run_details

  3. Summarize performance

    The agent looks at the loss and accuracy numbers to explain how the training is trending.

    get_run_details

  4. Report findings

    The agent presents the configuration and metric data in a clear format within your chat window.

    get_run_details

Try it

Copy these to start.

Use these prompts to begin interacting with your experiment data.

Starting points

Replace the project names and run IDs with your actual Weights & Biases identifiers.

Weights & Biases Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_I1Yu40vJsqaxsdrmMLW0Zf84kGzji4jYFvnrqvBH/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — Weights & Biases capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "weights-biases-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_I1Yu40vJsqaxsdrmMLW0Zf84kGzji4jYFvnrqvBH/mcp"
    }
  }
}
Copy into chat04

  • List all the recent runs in my 'image-classification' project.

  • What were the accuracy and loss values for run 'abc-123'?

  • Show me the configuration settings used in the 'resnet-v2' project runs.

  • Compare the loss curves for the last three runs in my 'nlp-finetuning' project.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect Weights & Biases once, then ask.

One connection gives your agent access to all your experiment data. Your credentials stay encrypted and secure.

Connect Weights & Biases to your AI

FAQ

How this task behaves

  • 01

    Can my AI change my training hyperparameters?

    No. The agent can only read data from existing runs. You must manually update your code or config files to change training behavior.

  • 02

    Can the AI delete old experiment runs?

    No. The agent is restricted to reading run details and listing runs. It cannot delete or modify any data in Weights & Biases.

  • 03

    Does the AI see all my projects?

    The agent can see any project your account has access to, but it only looks at the specific projects you mention in your prompt.

  • 04

    Can I ask for specific charts?

    The agent pulls the raw metric values. It can describe the trends and numbers, but it does not render interactive W&B charts inside the chat.

  • 05

    How does the agent know which run I'm talking about?

    You can provide the specific run ID, or you can ask the agent to list runs in a project so you can pick one.

  • More questions about Weights & Biases? The Connector page answers them. See everything the Weights & Biases Connector can do

Connect Weights & Biases to Claude, Cursor, ChatGPT & more