ReasonForge

Exact math for LLMs. No hallucinations.

ReasonForge is a modular mathematics toolkit that integrates with Claude and other LLMs via MCP (Model Context Protocol). Instead of letting the LLM guess at calculations, ReasonForge routes mathematical operations through SymPy, NumPy, and SciPy — returning provably correct results.

111 tools. 7 domain-specific servers. Zero probabilistic math.

The Problem

Ask an LLM to solve a differential equation, simplify a complex expression, or find a sequence pattern, and you'll get plausible-looking answers that are wrong 15-30% of the time. For research, finance, or engineering applications, "usually right" isn't good enough.

The Solution

ReasonForge gives your LLM access to a real computation engine. The LLM handles natural language understanding and problem decomposition; ReasonForge handles the actual math.

User: "Find the pattern in [1, 4, 9, 16, 25] and give me the equation"

Claude + ReasonForge:
  → Calls find_sequence_pattern tool
  → Returns: n² (exact symbolic formula, not a guess)

Quick Start

pip install reasonforge-logic

Or install the complete ecosystem:

pip install reasonforge[all]

Or install from source:

git clone https://github.com/foxintheloop/ReasonForge.git
cd ReasonForge
pip install -e packages/reasonforge -e packages/reasonforge-logic

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "reasonforge-logic": {
      "command": "python",
      "args": ["-m", "reasonforge_logic"]
    }
  }
}

Done. Claude now has access to 13 symbolic reasoning tools.

Example Queries

• "Convert (A OR B) AND C to conjunctive normal form"
• "Prove that a + b = b + a using commutativity"
• "Solve the system: 2x + y = 5, x - y = 1"
• "Find the derivative of x³ sin(x)"

Architecture

Install only what you need:

ReasonForge Ecosystem
│
├── reasonforge-core          # Pure Python symbolic engine (no MCP)
│
├── reasonforge-logic         # 13 tools ✅ Production Ready
│   └── Symbolic reasoning, formal logic, knowledge systems
│
├── reasonforge-expressions   # 16 tools (Beta)
│   └── Variable management, expression operations, basic calculus
│
├── reasonforge-algebra       # 18 tools (Beta)
│   └── Equation solving, matrices, optimization
│
├── reasonforge-analysis      # 17 tools (Beta)
│   └── Differential equations, transforms, signal processing
│
├── reasonforge-geometry      # 15 tools (Beta)
│   └── Vector/tensor calculus, coordinate systems
│
├── reasonforge-statistics    # 16 tools (Beta)
│   └── Probability, distributions, hypothesis testing
│
└── reasonforge-physics       # 16 tools (Beta)
    └── Classical mechanics, quantum states, electromagnetism

Why modular? A monolithic server loads ~8,000 tokens of tool definitions into context. A single package loads ~800-1,200 — an 85-90% reduction.

Package Status

Package	Tools	Status	Use Case
reasonforge-logic	13	✅ Ready	Formal proofs, boolean logic, pattern recognition
reasonforge-expressions	16	Beta	Expression simplification, substitution, basic calculus
reasonforge-algebra	18	Beta	Linear systems, polynomial roots, matrix operations
reasonforge-analysis	17	Beta	ODEs, PDEs, Laplace/Fourier transforms
reasonforge-geometry	15	Beta	Vectors, tensors, coordinate transforms
reasonforge-statistics	16	Beta	Distributions, inference, regression
reasonforge-physics	16	Beta	Mechanics, E&M, quantum computing

Note: Only reasonforge-logic is fully tested for production. Other packages are functional but may have edge cases.

Design Goals

Goal	Status
Exact symbolic computation	✅ Implemented via SymPy
Deterministic results	✅ Achieved for supported operations
Reproducibility	✅ Same input → same output
Full domain coverage	In progress

Documentation

• Usage Examples — 110 examples across all packages
• Individual package READMEs in /packages/

Contributing

# Run all tests
pytest packages/

# Run specific package tests
pytest packages/reasonforge-logic/tests/ -v

PRs welcome for new tools, packages, or docs.

License

MIT — see LICENSE

Citation

@software{reasonforge,
  title = {ReasonForge: Mathematics Toolkit for LLM Integration},
  author = {Derek Fox},
  year = {2025},
  url = {https://github.com/foxintheloop/ReasonForge}
}

---

Built on SymPy, NumPy, SciPy, and the Model Context Protocol.