Colabfit MCP Server

An MCP server that provides access to the ColabFit database

Local serverstdioPython

What is the Colabfit MCP server?

Most developer tooling work still happens through a UI a human drives. Colabfit MCP server moves it into the conversation instead. An MCP server that provides access to the ColabFit database.

The short version

This is a Model Context Protocol (MCP) server that gives AI assistants the ability to: - Search and download scientific datasets from ColabFit - Train MACE interatomic potentials on your local hardware (GPU or CPU) - Run energy/forces calculations and validate models with OpenKIM test drivers

Getting it running

mcp-cli on PyPI is all you need. Most clients run it directly, so configuration is a few lines and a restart.

The tools it exposes

The server publishes 10 tools. What each one is for:

  • search_datasets — Search ColabFit database by text, elements, properties, software
  • check_local_datasets — Scan local data directory for downloaded datasets, filter by elements/properties
  • download_dataset — Download a dataset from HuggingFace via KLIFF
  • train_mace — Train a MACE-style KLAY model from scratch using KLIFF
  • use_model — Run energy/forces/relax calculations with a trained KLAY model, or generate a Python snippet
  • check_status — Check GPU, packages, disk, existing models and datasets
  • list_test_drivers — List available kimvv test drivers, optionally filtered by property keyword
  • run_test_driver — Run a kimvv test driver against a trained KLAY model; saves structures.extxyz + results.json in a timestamped subdirectory; supports multiple
  • check_test_driver_result — Check status of an async test driver job and return inline results when complete
  • Requirements — The Requirements tool exposed by this server

What it needs from you

Configuration is passed through the environment: COLABFIT_DATA_ROOT, USER_ID, GROUP_ID, COLABFIT_BASE_URL. Treat anything key-shaped as a real credential — scope it to the minimum the server needs, and rotate it if it ever lands in a shared config.

  • Docker and Docker Compose v2 — for the containerized server - Git — for cloning the repository - make — for the quick-start commands (optional; manual steps are documented below) - (Optional) NVIDIA GPU + drivers — for GPU-accelerated training - (Optional) nvidia-container-toolkit

Things to watch

  • It runs with your machine's permissions. That is convenient and also the reason to think about what you point it at before you approve a tool call.
  • With 10 tools registered it takes up a noticeable share of the context window; turn it off in projects that never touch Colabfit.
  • Missing credentials fail quietly in some clients — if no tools show up, check the environment block first.
  • Keep per-call confirmation enabled while you learn its behaviour; it is the cheapest safeguard you have.

How it compares

Plenty of developer tooling servers cover similar ground. The differences that matter in practice are scope of access and how much setup stands between you and a working tool call. Colabfit's toolset — search_datasets, check_local_datasets, download_dataset and 7 more — is a fair guide to whether it matches your workflow. It is maintained by colabfit; worth a glance at recent repository activity before you build anything load-bearing on it.

SyncDev reviews every entry in this directory against the project's own documentation before publishing, and revisits them as servers change.

Available tools

ToolWhat it does
search_datasetsSearch ColabFit database by text, elements, properties, software
check_local_datasetsScan local data directory for downloaded datasets, filter by elements/properties
download_datasetDownload a dataset from HuggingFace via KLIFF
train_maceTrain a MACE-style KLAY model from scratch using KLIFF
use_modelRun energy/forces/relax calculations with a trained KLAY model, or generate a Python snippet
check_statusCheck GPU, packages, disk, existing models and datasets
list_test_driversList available kimvv test drivers, optionally filtered by property keyword
run_test_driverRun a kimvv test driver against a trained KLAY model; saves structures.extxyz + results.json in a timestamped subdirectory; supports multiple structures per call with optional repeat for supercell sizing and async_mode f
check_test_driver_resultCheck status of an async test driver job and return inline results when complete
RequirementsThe Requirements tool exposed by this server.

How to install the Colabfit MCP server

{
  "mcpServers": {
    "colabfit": {
      "command": "uvx",
      "args": ["mcp-cli"],
      "env": {
        "COLABFIT_DATA_ROOT": "your-value",
        "USER_ID": "your-value",
        "GROUP_ID": "your-value",
        "COLABFIT_BASE_URL": "your-value"
      }
    }
  }
}

Add to claude_desktop_config.json, then restart Claude Desktop.

Configuration

  • Docker and Docker Compose v2 — for the containerized server - Git — for cloning the repository - make — for the quick-start commands (optional; manual steps are documented below) - (Optional) NVIDIA GPU + drivers — for GPU-accelerated training - (Optional) nvidia-container-toolkit
VariableDescriptionRequired
COLABFIT_DATA_ROOTConfiguration value read at startup.Optional
USER_IDConfiguration value read at startup.Optional
GROUP_IDConfiguration value read at startup.Optional
COLABFIT_BASE_URLEndpoint or connection string the server talks to.Yes

Example prompts to try

  • Use Colabfit to search datasets.
  • Use Colabfit to check local datasets.
  • Use Colabfit to download dataset.

Frequently asked questions

It connects Colabfit to MCP-compatible AI assistants such as Claude and Cursor, exposing 10 tools (search_datasets, check_local_datasets, download_dataset, and more) that the assistant can call on your behalf. Instead of copying data back and forth by hand, the assistant works with Colabfit directly.