Trackio as an API and MCP Server

The Trackio dashboard can be configured to run as both an API server and an MCP (Model Context Protocol) server, allowing external tools and applications to programmatically interact with your experiment tracking data, or for you to be able to “chat with your experiment data” in natural language using ChatGPT, Claude, Deepseek, or various LLMs that support MCPs.

Setup

To enable API/MCP usage, start the MCP server when you launch the Trackio dashboard. You can do this a few different ways:

Option 1: CLI Command

trackio show --mcp-server

Option 2: Python Function

import trackio
trackio.show(mcp_server=True)

Option 3: Environment Variable

export GRADIO_MCP_SERVER=True
trackio show

When MCP server mode is enabled, Trackio will:

• Enable the Gradio API endpoints
• Start an MCP server alongside the web dashboard
• Allow programmatic access to writing and reading experiment data

API Usage

Once your Trackio server is running in MCP mode, click on the “Use via API or MCP” link in the footer of the dashboard to see the complete API documentation with exact usage examples for all endpoints.

Here is example usage in Python, JS, Bash, and via MCP.

from gradio_client import Client

# Connect to your Trackio instance
client = Client("http://127.0.0.1:7860/")

# Get all projects
result = client.predict(api_name="/get_all_projects")
print("Projects:", result)

# Get runs for a specific project
result = client.predict(
    project="my-project",
    api_name="/get_runs_for_project"
)
print("Runs:", result)

# Get metrics for a specific run
result = client.predict(
    project="my-project",
    run="run-1",
    api_name="/get_metrics_for_run"
)
print("Metrics:", result)

# Get metric values over time
result = client.predict(
    project="my-project",
    run="run-1",
    metric_name="loss",
    api_name="/get_metric_values"
)
print("Loss values:", result)

# Get project summary
result = client.predict(
    project="my-project",
    api_name="/get_project_summary"
)
print("Project summary:", result)

# Get run summary
result = client.predict(
    project="my-project",
    run="run-1",
    api_name="/get_run_summary"
)
print("Run summary:", result)

MCP Usage

When running as an MCP server, Trackio exposes its API endpoints as MCP tools that can be used by MCP-compatible clients like Claude Code.

MCP Server URL

The MCP server, using Streamble HTTP, is available (by default) at:

http://127.0.0.1:7860/gradio_api/mcp/

Available MCP Tools

The main tools that are available when Trackio is running in MCP server mode are:

1. get_all_projects - Get all project names. Returns a list of project names.
2. get_runs_for_project - Get all runs for a given project. Returns a list of run names.
3. get_metrics_for_run - Get all metrics for a given project and run. Returns a list of metric names.
4. get_metric_values - Get all values for a specific metric in a project/run. Returns a list of dictionaries with timestamp, step, and value.
5. get_project_summary - Get a summary of a project including number of runs and recent activity. Returns: Dictionary with project summary information.
6. get_run_summary - Get a summary of a specific run including metrics and configuration. Returns: Dictionary with run summary information.
7. bulk_log - Log metrics data to Trackio. Each entry is a dictionary with project, run, metrics, and optionally step and config.
8. upload_db_to_space - Upload database file to Hugging Face Space. Requires file upload capability.
9. bulk_upload_media - Upload media files for experiments. Requires file upload capability.

MCP Client Configuration

To add this MCP server to clients that support Streamable HTTP, add the following configuration to your MCP config:

{
  "mcpServers": {
    "gradio": {
      "url": "http://127.0.0.1:7860/gradio_api/mcp/"
    },
    "upload_files_to_gradio": {
      "command": "uvx",
      "args": [
        "--from",
        "gradio[mcp]",
        "gradio",
        "upload-mcp",
        "http://127.0.0.1:7860/gradio_api/mcp/",
        "<UPLOAD_DIRECTORY>"
      ]
    }
  }
}

The upload_files_to_gradio tool uploads files from your local UPLOAD_DIRECTORY to the Gradio app, which is needed for file-based operations. This tool requires uv to be installed. You can omit this tool if you are not planning on using the file-based tools.