Calc MCP Server

Calc MCP

📖 Documentation — Full tool reference, examples, and install guides.

21 tools for things AI is not good at — deterministic math, cryptographic randomness, accurate date arithmetic, encoding, hashing, and more.

LLMs hallucinate calculations, can't generate true random numbers, and struggle with timezones. This MCP server fixes that.

Quick Start

# Claude Code
claude mcp add -s user calc-mcp -- npx --prefix /tmp -y @coo-quack/calc-mcp@latest

# Or just run it
npx --prefix /tmp -y @coo-quack/calc-mcp@latest

Works with Claude Desktop, VS Code Copilot, Cursor, Windsurf, and any MCP client — setup guides below.

---

Why?

• Deterministic — Same input, same correct output, every time
• Secure — Sandboxed math, ReDoS protection, weak hash warnings
• Private — All computation runs locally, no data sent to external services
• No server config — Install once via npx; MCP client setup required
• No API key — No account or API key required for calc-mcp itself; requires Node.js

Examples

Ask in natural language — your AI assistant selects the appropriate tool.

Math & Numbers

Text & Encoding

Date & Time

Generation

Conversion

Analysis & Parsing

Decode & Parse

All 21 Tools

Install

Claude Code

claude mcp add -s user calc-mcp -- npx --prefix /tmp -y @coo-quack/calc-mcp@latest

Claude Desktop / Cursor / Windsurf

Add to your config file:

{
  "mcpServers": {
    "calc-mcp": {
      "command": "npx",
      "args": ["--prefix", "/tmp", "-y", "@coo-quack/calc-mcp@latest"]
    }
  }
}

VS Code (GitHub Copilot)

Add to .vscode/mcp.json in your workspace:

{
  "servers": {
    "calc-mcp": {
      "command": "npx",
      "args": ["--prefix", "/tmp", "-y", "@coo-quack/calc-mcp@latest"]
    }
  }
}

Docker Image

Calc MCP is available as a Docker image from GitHub Container Registry:

docker run --rm -i ghcr.io/coo-quack/calc-mcp:latest

Or use in MCP client config:

{
  "mcpServers": {
    "calc-mcp": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "ghcr.io/coo-quack/calc-mcp:latest"]
    }
  }
}

Available tags:

• ghcr.io/coo-quack/calc-mcp:latest — Latest release
• ghcr.io/coo-quack/calc-mcp:X.Y.Z — Specific version (replace X.Y.Z with the desired version)

Other MCP Clients

Calc MCP works with any MCP-compatible client. Run the server via stdio:

npx --prefix /tmp -y @coo-quack/calc-mcp@latest

Point your client's MCP config to the command above. The server communicates over stdio using the standard Model Context Protocol.

Development

bun install
bun run dev       # Start dev server
bun test          # Run tests
bun run lint      # Biome
bun run format    # Biome

Security

calc-mcp processes all data locally and does not:

• ❌ Send data to external servers
• ❌ Log data to files or remote services
• ❌ Store processed data persistently

For detailed security information, see SECURITY.md.

Safe Usage with LLMs

calc-mcp itself is local-only. However, when used via an LLM, your inputs are sent to the LLM provider (Anthropic, OpenAI, etc.).

• ✅ DO: Use test/sample data when possible
• ✅ DO: Use local-only LLMs for sensitive operations
• ❌ DON'T: Pass production secrets to MCP tools (they will be sent to your LLM provider)

Example:

# ❌ Unsafe: Any secret passed to MCP tool is sent to your LLM provider
# Tool: hash
# Input: { "input": "sk-1234567890abcdef", "algorithm": "sha256" }

# ✅ Safe: Use test data only (for learning/development)
# Tool: hash
# Input: { "input": "test-value-123", "algorithm": "sha256" }

# ✅ For production secrets: Use local-only LLMs or process outside MCP

Note: Error messages are automatically sanitized to prevent accidental data leakage.

For security issues, please see our Security Policy.

License

MIT