MCP-ORTools

A Model Context Protocol (MCP) server implementation using Google OR-Tools for constraint solving. Designed for use with Large Language Models through standardized constraint model specification.

Overview

MCP-ORTools integrates Google's OR-Tools constraint programming solver with Large Language Models through the Model Context Protocol, enabling AI models to:

• Submit and validate constraint models

• Set model parameters

• Solve constraint satisfaction and optimization problems

• Retrieve and analyze solutions

Installation

1. Install the package:

2. Configure Claude Desktop Create the configuration file at %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

Model Specification

Models are specified in JSON format with three main sections:

• variables: Define variables and their domains

• constraints: List of constraints using OR-Tools methods

• objective: Optional optimization objective

Constraint Syntax

Constraints must use OR-Tools method syntax:

• .__le__() for less than or equal (<=)

• .__ge__() for greater than or equal (>=)

• .__eq__() for equality (==)

• .__ne__() for not equal (!=)

Usage Examples

Simple Optimization Model

Knapsack Problem

Example: Select items with values [3,1,2,1] and weights [2,2,1,1] with total weight limit of 2.

Additional constraints example:

Features

• Full OR-Tools CP-SAT solver support

• JSON-based model specification

• Support for:

  • Integer and boolean variables (domain: [min, max])

  • Linear constraints using OR-Tools method syntax

  • Linear optimization objectives

  • Timeouts and solver parameters

  • Binary constraints and relationships

  • Portfolio selection problems

  • Knapsack problems

Supported Operations in Constraints

• Basic arithmetic: +, -, *

• Comparisons: .le(), .ge(), .eq(), .ne()

• Linear combinations of variables

• Binary logic through combinations of constraints

Development

To setup for development:

Model Response Format

The solver returns solutions in JSON format:

Status values:

• OPTIMAL: Found optimal solution

• FEASIBLE: Found feasible solution

• INFEASIBLE: No solution exists

• UNKNOWN: Could not determine solution

License

MIT License - see LICENSE file for details