MCP Server for Scientific Calculation

A Model Context Protocol (MCP) server that leverages Large Language Models (LLMs) to perform and assist in scientific calculations.

📦 Requirement

We successfully test the code with softwares as the followings:

• Julia = 1.11.6
  • ITensors = "0.9.9"
  • ITensorMPS = "0.3.19"
  • HDF5 = "0.17.2"
  • CSV.jl = "0.10.15"
  • DataFrames.jl = "1.7.0"
• Python = 3.13.5, initialized by uv, with uuid and juliacall installed.

🚀 Quick Start

Clone the repository and install dependencies:

git clone https://github.com/xing-phys/MCP_TN.git

Precompile the Julia package to get a system image TenKits.so:

cd MCP_TN/TenKits
julia generate_so.jl

Configure the server script

Edit the main.py file:

vim MCP_TN/main.py

Replace the value of os.environ['PYTHON_JULIACALL_SYSIMAGE'] to the actual local path to TenKits.so

Go to the followings to see how to use in an LLM client

🖥️ Integration with LLM Clients

Open an LLM Client (Claude Desktop, 5ire, etc) and configure the MCP server:

• Sever name: MCP_TN
• Command: uv
• Arguments: run mcp run /path/to/your/MCP_TN/main.py (Replace with your actual local path to main.py)
• Environment:
  • Variable: UV_PROJECT
  • Value: /path/to/your/MCP_NetKet (Replace with your actual local path to the MCP_TN folder)

🎯 Interactive Demo: Ferromagnetic XXZ Heisenberg Model

Try this step-by-step demo to see the MCP server in action! This demonstrates a work flow for calculate the ground state magnetization of the XXZ model.

Step 1: Generate the MPO of the Hamiltonian

Create the MPO of this XXZ model:
$$
H = \sum_{j=1}^{N-1} J(S^{x}_{j}S^{x}_{j+1} + S^{y}_{j}S^{y}_{j+1} ) + J^{z} S^{z}_{j}S^{z}_{j+1}
$$
set $J = -1$, $J^{z}$ from -1.5 to -0.5, step 0.1. The system size is set as $N = 50$.

Expected Results: The MPO for each $J^{z}$ is generated properly and saved to the desired /tmp/ folder and labeled by an unique UUID.

Step 2: Ground State Calculation

For each Jz, calculate the ground state

Expected Results: The ground energies are returned. The ground state MPS for each $J^{z}$ is generated properly and saved to the desired /tmp/ folder and labeled by an unique UUID.

Step 3: Calculate the Magnetization

Calculate the average magnetization $J^{z}$
$$
\sum_{j=1}^{N} \frac{\left<S^{z}_{j}\right>}{L}
$$

Expected Results: Get the average magnetization with absolute values about 0.5 for $J^{z} < -1$ and nearly 0 for $J^{z} > -1$.

📂 Project Structure

MCP_TN
│── README.md
│── TenKits
│   │── Manifest.toml
│   │── Project.toml
│   │── generate_so.jl
│   │── precompile_script.jl
│   └── src
│       └── TenKits.jl
│── __pycache__
│   └── main.cpython-313.pyc
│── main.py
│── pyproject.toml
└── uv.lock