Atomistic Toolkit MCP Server

What is Atomistic Toolkit MCP Server?

An MCP (Model Context Protocol) server for atomistic modeling workflows, powered by ASE, pymatgen, and modern machine‑learning interatomic potentials. It exposes ready‑to‑call tools for building structures, running geometry optimization and molecular dynamics, analyzing structures and trajectories, and providing downloadable file artifacts.

How to use Atomistic Toolkit MCP Server?

Requires Python 3.11+. Install dependencies with pip install -r requirements.txt, then start the HTTP server with uvicorn mcp_atomictoolkit.http_app:app --host 0.0.0.0 --port 10000 (or python main.py). For desktop MCP clients, use STDIO mode: python -m mcp_atomictoolkit.mcp_server. Verify with curl -s http://localhost:10000/healthz.

Key features of Atomistic Toolkit MCP Server

• MCP‑native workflows via FastMCP tools
• Structure generation: bulk, surface, molecule, supercell, amorphous, liquid, bicrystal, polycrystal
• Optimization with MLIPs (kim default, nequip, orb)
• Molecular dynamics (Velocity Verlet, Langevin, NVT Berendsen)
• Analysis: RDF, MSD, VACF (Green‑Kubo), thermodynamic trends
• Downloadable artifacts (xyz, extxyz, cif, traj, png, svg, csv, dat)

Use cases of Atomistic Toolkit MCP Server

• Automate atomistic simulation workflows without hand‑writing scripts
• Run structure building, optimization, and MD from LLM‑based agents or desktop clients
• Analyze trajectories for diffusion coefficients and correlation functions
• Generate and share simulation artifacts (files + plots) via download URLs

FAQ from Atomistic Toolkit MCP Server

What modes does the server support?

It supports both Streamable HTTP (via uvicorn) and STDIO (for desktop MCP clients). STDIO mode must keep stdout clean of print/logging.

What file formats are available for download?

Artifacts can be downloaded in .xyz, .extxyz, .cif, .traj, .png, .svg, .csv, .dat, and more.

What endpoints does the server expose?

Primary MCP endpoint is POST /. Also includes GET /healthz, GET /docs, GET /.well-known/mcp/server-card.json, GET /artifacts/{artifact_id}/{filename}, and /sse/.

Can I deploy the server in a container or on Render?

Yes. A render.yaml and a Dockerfile are provided. Default Docker command runs on port 7860.

What optimization and MD parameters can be tuned?

Optimization supports max_steps, fmax, maxstep, alpha, and constraints (fixed atoms, bonds, cell). MD supports temperature_K, friction, taut, and multiple integrators (NVE, NVT‑Langevin, NVT‑Berendsen).