How it works
Parsing pipeline
ShapeMCP uses purpose-built parsers for each input type: the TypeScript Compiler API for TS/JS, SQL parsers for DDL, GraphQL parsers for SDL, and structured parsers for JSON and .env. It identifies boundaries (e.g. exported functions, table definitions, query fields) and extracts names, types, descriptions, and required flags.
Mapping to MCP
Extracted metadata is mapped directly to the JSON Schema structures required by the Model Context Protocol: name, description, and inputSchema with properties and required. The output is valid for use in Claude, Cursor, and any MCP-compliant client.
Where it runs
In the browser, parsing and conversion run entirely client-side. On a Vercel (or similar) deployment, the same logic runs on the server via API routes. Your code never has to leave your infrastructure unless you choose to call a hosted API.