Colabfit MCP

Install

Works with Claude Code, Codex, VS Code, Cursor, Windsurf, Claude Desktop. Read the repository before you give any tool access to your data.

npx agentmarket install connector/colabfit-colabfit-mcp

About

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this server can do and what it asks for, from its tool list and install config. Checked Oct 25, 2026.

Tools

2 read · 1 write · 0 delete · 1 not yet classified

Secrets

None asked for

Install scripts

Not scanned yet

Runs as

Python package

Tools (4)

Read from the source code; the running server may differ. Descriptions are the server’s own words; hidden characters are shown as [U+…].

• build_dataset

  Build a custom filtered dataset by querying ColabFit VastDB across all datasets. Filters configurations by computation method, software, formula, required properties, and elements present. Exports to extxyz format ready for train_mace. Re-running with the same dataset_name returns cached results. Use a different dataset_name to regenerate with new filters. methods_contain: Substring filters on method names (case-insensitive). "DFT" matches "DFT-PBE", "DFT+U", etc. software_contain: Substring filters on software names (case-insensitive). "VASP" matches "VASP 6.3.2", etc. dataset_ids:…

• check_status

  Check system status including GPU, packages, and connectivity. Returns versions of key packages (torch, klay, kliff), GPU availability, disk usage, and lists of existing models and datasets. IMPORTANT: All file paths are inside the Docker container filesystem. Use container_paths values when specifying paths to other tools. Returns: Dict with system info, connectivity, resource inventory, and container_paths showing where datasets/models/outputs are stored.

• search_datasets

  Search ColabFit database for materials science datasets. ## DISCOVERY PROTOCOL — ALWAYS FOLLOW THIS ORDER 1. Call check_local_datasets first to find already-downloaded datasets in the Docker container (/home/mcpuser/colabfit/datasets/). If a suitable match exists locally, skip steps 2-3 and proceed directly to train_mace. 2. If no local match, call this tool to search ColabFit.org. ColabFit is the authoritative source — do NOT search the internet or guess dataset names. 3. Download the chosen dataset with download_dataset, then train with train_mace. ## FILTERING BY DATASET SIZE — …

• train_mace

  Train a MACE-style KLAY model using KLIFF on XYZ data. IMPORTANT: All file paths are inside the Docker container filesystem. Datasets are under /home/mcpuser/colabfit/datasets/. Trained models are saved under /home/mcpuser/colabfit/models/. ## DATASET SELECTION — use exactly one of: dataset_name, train_file, or elements Priority order: train_file > dataset_name > elements/auto-discovery. dataset_name (PREFERRED when you have a specific dataset): Exact folder name of a locally downloaded dataset, as returned by download_dataset (e.g. dataset_name="mlearn_Ni_test"). Performs a di…

Signals

Stars

2

Forks

1

Open issues

0

Weekly downloads

31

Last commit

Apr 23, 2026 (6 months ago)

Last release

Oct 16, 2025 (1 year ago)

Last package publish

Oct 16, 2025 (1 year ago)

From GitHub (fetched Oct 24, 2026) and the package registry.

Details

Registry name

io.github.colabfit/colabfit-mcp

Version

1.0.0

Auth

No auth

Packages

pypi colabfit-mcp

License

License unknown

Repository

github.com/colabfit/colabfit-mcp

First seen

Oct 24, 2026

Last checked

Oct 25, 2026

Source

Official MCP Registrygithub-graphql