Calculator

Exposes mathematical expression evaluation through the Model Context Protocol (MCP) using a standardized JSON-RPC 2.0 interface. The system registers a 'calculate' tool within the MCP framework that accepts string expressions and returns computed results, enabling LLM clients to invoke calculations through a protocol-agnostic communication layer rather than direct function calls. FastMCP framework handles protocol marshaling, request routing, and response serialization automatically.

Evaluates mathematical expressions in a restricted execution environment that whitelists only safe mathematical functions (arithmetic operators, trigonometry, logarithms, etc.) while blocking dangerous operations like file I/O, system calls, or arbitrary code execution. The expression evaluator uses a security model that validates input syntax before execution and restricts the namespace available to eval() to a curated set of math functions from Python's math module, preventing injection attacks and unintended side effects.

Provides access to Python's standard math module functions (trigonometric: sin, cos, tan; logarithmic: log, log10, log2; exponential: exp, sqrt; constants: pi, e; and others) through the sandboxed expression evaluator. These functions are pre-imported into the evaluation namespace, allowing expressions like 'sin(pi/2)' or 'sqrt(16)' to execute without explicit imports. The binding is static — the set of available functions is fixed at server startup and cannot be extended at runtime.

Validates mathematical expressions for syntax errors before execution and returns detailed error messages when evaluation fails. The system catches exceptions during expression evaluation (SyntaxError, NameError, TypeError, ZeroDivisionError, etc.) and returns human-readable error descriptions to the LLM client, enabling the LLM to correct malformed expressions and retry. Error messages include the type of error and context about what went wrong, facilitating debugging of LLM-generated expressions.

Packages the calculator as a deployable MCP server that runs as an independent process communicating with MCP clients via JSON-RPC over stdio or network sockets. Supports two installation methods: uvx (direct execution without local installation) and pip (traditional Python package installation). The server bootstraps via a main() entry point that initializes the FastMCP framework, registers the calculate tool, and enters the MCP protocol event loop, handling incoming client requests until shutdown.

Runs on Linux, macOS, and Windows with only Python 3.10+ and the mcp library as runtime dependencies, requiring no system-specific compilation or platform-specific code paths. The codebase uses only standard library modules (math, json, sys) and the mcp framework, avoiding heavy dependencies like numpy or scipy that require compilation. This minimal dependency footprint enables rapid deployment across heterogeneous environments and reduces supply chain risk.