USolver

How to Set Up and Use USolver

USolver is a Python-based MCP server that implements a universal solver interface using the 'LLM sandwich' model, where natural language problem descriptions are translated into mathematical constraints, passed to dedicated solver backends (Google OR-Tools for combinatorial optimization, CVXPy for convex optimization, and Z3 for SMT/logical constraint satisfaction), and the results are verbalized back in plain language. It is designed for engineers, operations researchers, and data scientists who want to describe optimization problems — staff scheduling, pipeline design, resource allocation — in conversational language rather than solver-specific syntax, with the LLM handling the translation layer in both directions.

git clone https://github.com/sdiehl/usolver.git
cd usolver

# Using uv (recommended):
uv sync

# Or using pip:
pip install google-or-tools cvxpy z3-solver

python install.py

{
  "mcpServers": {
    "usolver": {
      "command": "python",
      "args": ["/path/to/usolver/main.py"]
    }
  }
}

{
  "mcpServers": {
    "usolver": {
      "command": "python",
      "args": ["/path/to/usolver/main.py"]
    }
  }
}

- "Schedule 4 nurses across 3 shifts (morning 6-14, afternoon 14-22, night 22-6) so each nurse works 5 days and shifts are evenly distributed"
- "Design a water transport pipeline with max flow rate 500 L/min, pressure between 2 and 8 bar, and pipe diameter between 50 and 200mm — minimize cost"
- "Find integer values of x and y such that 3x + 2y = 18, x > 0, y > 0, and x + y is minimized"
- "Is it possible to color a map with 3 colors such that no two adjacent regions share a color? Regions: A-B, B-C, C-A, A-D"