BrainyCalc Docs — Overview
Docs / Overview
Documentation

BrainyCalc Overview

BrainyCalc is a Model Context Protocol (MCP) math server that gives LLMs and AI agents accurate, independently verified, and reproducible mathematical results — with full auditability. Instead of letting an LLM estimate mathematical answers, the AI delegates computation to BrainyCalc's deterministic engine.

At a glance

Reproducibility
100%
Same input, same output
Error bound
±10⁻¹⁵
Per computation
Verification
6 steps
Every result checked
Audit trail
Full
Cryptographic proof

Quick Start

Add BrainyCalc to any MCP-compatible agent in one step. Add the following block to your MCP configuration file:

claude_desktop_config.json / mcp.json
{
  "mcpServers": {
    "brainycalc": {
      "command": "npx",
      "args": ["-y", "brainycalc"]
    }
  }
}
Node.js required

BrainyCalc requires Node.js ≥ 18 and npm ≥ 7. The npx -y flag automatically downloads the latest version on first run — no separate install step needed.

How it works

When your AI agent encounters a computation, it sends a tool call to BrainyCalc over standard MCP JSON-RPC stdio. BrainyCalc processes it through a 6-step pipeline and returns a verified, auditable result.

Compute

Uses reliable numerical or symbolic methods. Never estimates.

Verify

An independent second pass catches instabilities before results reach your agent.

Reproduce

Same input always produces the same output — non-determinism is eliminated.

Report confidence

Every result includes a confidence score and error bound.

Record provenance

Engine version, method, and parameters captured with every result.

Audit trail

Fingerprinted and stored — inspect or reproduce any result at any time.

Tool categories

BrainyCalc exposes 100+ tools across four categories. Click any category to explore its full API reference.

Requirements

Dependency Minimum Version Notes
node 18 ESM module support required
npm 7 For npx invocation

Compatible platforms

BrainyCalc speaks standard MCP over JSON-RPC stdio and works with any MCP-compatible platform:

PlatformNotes
Claude Desktop (Anthropic)Add to claude_desktop_config.json
LangChain agentsUse the MCP adapter for LangChain
Custom SDK implementationsStandard JSON-RPC stdio transport
Self-hosted / cluster deploymentsLocally runnable for lower latency
Tool discovery

Tool schemas are automatically exposed via the MCP tools/list endpoint — no manual wiring required. Your agent discovers all 100+ tools on first connection.

© 2025 BrainyCalc. Trusted computation for AI.