Which integrations are available for this server?

Leverages Gradio's built-in MCP server support to expose trackio functionality as tools for AI agents Enables seamless deployment to Hugging Face Spaces, with support for both public and private spaces, and includes authentication for private datasets

How do I use trackio-mcp?

1. Click on "Install Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@trackio-mcp show me the latest metrics from my image-classification project" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

trackio-mcp

PyPI version License: MIT

MCP (Model Context Protocol) server support for trackio experiment tracking

This package enables AI agents to observe and interact with trackio experiments through the Model Context Protocol (MCP). Simply import trackio_mcp before trackio to automatically enable MCP server functionality.

Features

Zero-code integration: Just import trackio_mcp before trackio
Automatic MCP server: Enables MCP server on all trackio deployments (local & Spaces)
Rich tool set: Exposes trackio functionality as MCP tools for AI agents
Spaces compatible: Works seamlessly with Hugging Face Spaces deployments
Drop-in replacement: No changes needed to existing trackio code

Related MCP server: ActiveCampaign MCP Server

Installation

pip install trackio-mcp

Or with development dependencies:

pip install trackio-mcp[dev]

Quick Start

Basic Usage

Simply import trackio_mcp before importing trackio:

import trackio_mcp  # This enables MCP server functionality
import trackio as wandb

# Your existing trackio code works unchanged
wandb.init(project="my-experiment")
wandb.log({"loss": 0.1, "accuracy": 0.95})
wandb.finish()

The MCP server will be automatically available at:

Local: http://localhost:7860/gradio_api/mcp/sse
Spaces: https://your-space.hf.space/gradio_api/mcp/sse

Deploy to Hugging Face Spaces with MCP

import trackio_mcp
import trackio as wandb

# Deploy to Spaces with MCP enabled automatically
wandb.init(
    project="my-experiment", 
    space_id="username/my-trackio-space"
)

wandb.log({"loss": 0.1})
wandb.finish()

Standalone MCP Server

Launch a dedicated MCP server for trackio tools:

from trackio_mcp.tools import launch_trackio_mcp_server

# Launch standalone MCP server on port 7861
launch_trackio_mcp_server(port=7861, share=False)

Available MCP Tools

Once connected, AI agents can use these trackio tools:

Core Tools (via Gradio API)

log: Log metrics to a trackio run
upload_db_to_space: Upload local database to a Space

Extended Tools (via trackio-mcp)

get_projects: List all trackio projects
get_runs: Get runs for a specific project
filter_runs: Filter runs by name pattern
get_run_metrics: Get metrics data for a specific run
get_available_metrics: Get all available metric names for a project
load_run_data: Load and process run data with optional smoothing
get_project_summary: Get comprehensive project statistics

Example Agent Interaction

Human: "Show me the latest results from my 'image-classification' project"

Agent: I'll check your trackio projects and get the latest results.

[Tool: get_projects] → finds "image-classification" project
[Tool: get_runs] → gets runs for "image-classification" 
[Tool: get_run_metrics] → gets metrics for latest run
[Tool: get_available_metrics] → gets metric names

Agent: Your latest image-classification run achieved 94.2% accuracy with a final loss of 0.18. The model trained for 50 epochs with best validation accuracy of 94.7% at epoch 45.

MCP Client Configuration

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or equivalent:

Public Spaces:

{
  "mcpServers": {
    "trackio": {
      "url": "https://your-space.hf.space/gradio_api/mcp/sse"
    }
  }
}

Private Spaces/Datasets:

{
  "mcpServers": {
    "trackio": {
      "url": "https://your-private-space.hf.space/gradio_api/mcp/sse",
      "headers": {
        "Authorization": "Bearer YOUR_HF_TOKEN"
      }
    }
  }
}

Local Development:

{
  "mcpServers": {
    "trackio": {
      "url": "http://localhost:7860/gradio_api/mcp/sse"
    }
  }
}

See Claude Code MCP docs for more info.

Public Spaces:

claude mcp add --transport sse trackio https://your-space.hf.space/gradio_api/mcp/sse

Private Spaces/Datasets:

claude mcp add --transport sse --header "Authorization: Bearer YOUR_HF_TOKEN" trackio https://your-private-space.hf.space/gradio_api/mcp/sse

Local Development:

{
  "mcpServers": {
    "trackio": {
      "type": "sse",
      "url": "http://localhost:7860/gradio_api/mcp/sse"
    }
  }
}

Add to your Cursor ~/.cursor/mcp.json file or create .cursor/mcp.json in your project folder. See Cursor MCP docs for more info.

Public Spaces:

{
  "mcpServers": {
    "trackio": {
      "url": "https://your-space.hf.space/gradio_api/mcp/sse"
    }
  }
}

Private Spaces/Datasets:

{
  "mcpServers": {
    "trackio": {
      "url": "https://your-private-space.hf.space/gradio_api/mcp/sse",
      "headers": {
        "Authorization": "Bearer YOUR_HF_TOKEN"
      }
    }
  }
}

Local Development:

{
  "mcpServers": {
    "trackio": {
      "url": "http://localhost:7860/gradio_api/mcp/sse"
    }
  }
}

Add to your Windsurf MCP config file. See Windsurf MCP docs for more info.

Public Spaces:

{
  "mcpServers": {
    "trackio": {
      "serverUrl": "https://your-space.hf.space/gradio_api/mcp/sse"
    }
  }
}

Private Spaces/Datasets:

{
  "mcpServers": {
    "trackio": {
      "serverUrl": "https://your-private-space.hf.space/gradio_api/mcp/sse",
      "headers": {
        "Authorization": "Bearer YOUR_HF_TOKEN"
      }
    }
  }
}

Local Development:

{
  "mcpServers": {
    "trackio": {
      "serverUrl": "http://localhost:7860/gradio_api/mcp/sse"
    }
  }
}

Add to .vscode/mcp.json. See VS Code MCP docs for more info.

Public Spaces:

{
  "mcp": {
    "servers": {
      "trackio": {
        "type": "http",
        "url": "https://your-space.hf.space/gradio_api/mcp/sse"
      }
    }
  }
}

Private Spaces/Datasets:

{
  "mcp": {
    "servers": {
      "trackio": {
        "type": "http", 
        "url": "https://your-private-space.hf.space/gradio_api/mcp/sse",
        "headers": {
          "Authorization": "Bearer YOUR_HF_TOKEN"
        }
      }
    }
  }
}

Local Development:

{
  "mcp": {
    "servers": {
      "trackio": {
        "type": "http",
        "url": "http://localhost:7860/gradio_api/mcp/sse"
      }
    }
  }
}

Add to mcp.json in your project directory. See Gemini CLI Configuration for details.

Public Spaces:

{
  "mcpServers": {
    "trackio": {
      "command": "npx",
      "args": ["mcp-remote", "https://your-space.hf.space/gradio_api/mcp/sse"]
    }
  }
}

Private Spaces/Datasets:

{
  "mcpServers": {
    "trackio": {
      "command": "npx", 
      "args": ["mcp-remote", "https://your-private-space.hf.space/gradio_api/mcp/sse"],
      "env": {
        "HF_TOKEN": "YOUR_HF_TOKEN"
      }
    }
  }
}

Local Development:

{
  "mcpServers": {
    "trackio": {
      "command": "npx",
      "args": ["mcp-remote", "http://localhost:7860/gradio_api/mcp/sse"]
    }
  }
}

Create .cursor/mcp.json (or equivalent for your IDE):

Public Spaces:

{
  "mcpServers": {
    "trackio": {
      "url": "https://your-space.hf.space/gradio_api/mcp/sse"
    }
  }
}

Private Spaces/Datasets:

{
  "mcpServers": {
    "trackio": {
      "url": "https://your-private-space.hf.space/gradio_api/mcp/sse",
      "headers": {
        "Authorization": "Bearer YOUR_HF_TOKEN"
      }
    }
  }
}

Local Development:

{
  "mcpServers": {
    "trackio": {
      "url": "http://localhost:7860/gradio_api/mcp/sse"
    }
  }
}

Configuration

Environment Variables

TRACKIO_DISABLE_MCP: Set to "true" to disable MCP functionality (default: MCP enabled)

Programmatic Control

import os
os.environ["TRACKIO_DISABLE_MCP"] = "true"  # Disable MCP
import trackio_mcp  # MCP won't be enabled
import trackio

How It Works

trackio-mcp uses monkey-patching to automatically:

Enable MCP server: Sets mcp_server=True on all Gradio launches
Enable API: Sets show_api=True to expose Gradio API endpoints
Add tools: Registers additional trackio-specific MCP tools
Preserve compatibility: No changes needed to existing trackio code

The package patches:

gradio.Blocks.launch() - Core Gradio launch method
trackio.ui.demo.launch() - Trackio dashboard launches
Adds new MCP endpoints at /gradio_api/mcp/sse

Deployment Examples

Local Development

import trackio_mcp
import trackio

# Start local tracking with MCP enabled
trackio.show()  # Dashboard + MCP server at http://localhost:7860

Public Spaces Deployment

import trackio_mcp
import trackio as wandb

# Deploy to public Spaces with MCP support
wandb.init(
    project="public-model",
    space_id="username/model-tracking"
)

wandb.log({"epoch": 1, "loss": 0.5})
wandb.finish()

Private Spaces/Datasets Deployment

import trackio_mcp
import trackio as wandb

# Deploy to private Spaces with private dataset
wandb.init(
    project="private-model",
    space_id="organization/private-model-tracking",  # Private space
    dataset_id="organization/private-model-metrics"  # Private dataset
)

wandb.log({"epoch": 1, "loss": 0.5})
wandb.finish()

CLI Interface

# Launch standalone MCP server
trackio-mcp server --port 7861

# Check status and configuration
trackio-mcp status

# Test MCP server functionality
trackio-mcp test --url http://localhost:7860

Security Considerations

Private Spaces: Use HF tokens for authentication with private spaces/datasets
Access Control: MCP server inherits trackio's access controls
Network Security: Consider firewall rules for production deployments
Token Management: Store HF tokens securely, use environment variables

Troubleshooting

MCP Server Not Available

import trackio_mcp
import trackio

# Check if MCP was disabled
import os
print("MCP Disabled:", os.getenv("TRACKIO_DISABLE_MCP"))

# Manual verification
trackio.show()  # Look for MCP server URL in output

Connection Issues

Check URL: Ensure correct /gradio_api/mcp/sse endpoint
Authentication: Add Bearer token for private Spaces/datasets
Network: Verify firewall/proxy settings
Dependencies: Ensure gradio[mcp] is installed

Tool Discovery Issues

# Test tools manually
from trackio_mcp.tools import register_trackio_tools

tools = register_trackio_tools()
tools.launch(mcp_server=True)  # Test tools interface

Contributing

Fork the repository
Install development dependencies: pip install -e .[dev]
Make your changes
Run tests: pytest
Submit a pull request

License

MIT License - see LICENSE file.

Acknowledgments

trackio - The excellent experiment tracking library
Gradio - For built-in MCP server support
Model Context Protocol - For the standardized AI tool protocol

Made with care for the AI research community

trackio-mcp

trackio-mcp

Features

Installation

Quick Start

Basic Usage

Deploy to Hugging Face Spaces with MCP

Standalone MCP Server

Available MCP Tools

Core Tools (via Gradio API)

Extended Tools (via trackio-mcp)

Example Agent Interaction

MCP Client Configuration

Configuration

Environment Variables

Programmatic Control

How It Works

Deployment Examples

Local Development

Public Spaces Deployment

Private Spaces/Datasets Deployment

CLI Interface

Security Considerations

Troubleshooting

MCP Server Not Available

Connection Issues

Tool Discovery Issues

Contributing

License

Acknowledgments

Resources

Latest Blog Posts

MCP directory API