Trackio

You can enable and use the MCP (Model Context Protocol) server for trackio experiments with zero-code integration. By importing the MCP helper before trackio, you automatically expose trackio’s capabilities as MCP tools that AI agents can discover and execute, allowing you to observe and interact with your experiments from external agents or dashboards.

How to use

You enable MCP server functionality by importing trackio_mcp before you import trackio in your code. This activates the MCP server automatically for all trackio deployments, including local runs and Hugging Face Spaces.

Once enabled, AI agents can connect to the MCP server to access core and extended tools that let them log metrics, fetch project data, and query run details. The local MCP server is available at http://localhost:7860/gradio\_api/mcp/sse, and deployments to Spaces expose a similar endpoint under your space URL.

If you prefer a standalone MCP server dedicated to trackio tools, you can start it separately on port 7861. This provides an independent endpoint for MCP interactions.

How to install

Prerequisites: ensure you have Python and pip installed on your system.

Install the MCP package for trackio.

pip install trackio-mcp

Optional development dependencies.

pip install trackio-mcp[dev]

Use the drop-in approach in your code to enable MCP before trackio imports.

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

wandb.init(project="my-experiment")
wandb.log({"loss": 0.1, "accuracy": 0.95})
wandb.finish()

Available tools

log

Log metrics to a trackio run via the MCP interface.

upload_db_to_space

Upload the local trackio database to a Hugging Face Space.

get_projects

List all trackio projects available to the MCP client.

get_runs

Retrieve runs for a specified trackio project.

filter_runs

Filter runs by name patterns or criteria.

get_run_metrics

Fetch metrics data for a specific run.

get_available_metrics

List all metric names available for a project.

load_run_data

Load and process run data with optional smoothing.

get_project_summary

Return comprehensive statistics for a project.