What is aias-spec?

aias-spec turns ordinary annotated functions in your JavaScript or TypeScript code into a working MCP server, which is the kind of server that lets AI tools like Claude Desktop call your code as a tool. Instead of writing an OpenAPI spec by hand or building a server from scratch, you write special comments, starting with @ai, directly above a function describing what it does, what parameters it takes, and how the AI should confirm the action back to the user. Running the aias generate command reads those comments and produces the server code for you. The project claims a benchmark result showing that tools generated this way made no wrong tool-call errors across 25 test requests, and correctly refused to guess when a request was missing required information, compared to a plain conversion from an OpenAPI spec. To use it, you need Node.js version 20 or newer. After installing the package, you run a setup command to create a config file, then annotate your functions with the special comments and run the generate command. The tool checks your annotations carefully and will fail with a clear error, pointing to the exact file and line, if a parameter name does not match your real function, if a confirmation message references an unknown parameter, or if required information is missing. Functions marked private are never turned into callable tools. The tool currently understands three common ways of writing functions in JavaScript and TypeScript, including standalone exported functions, exported constants that hold functions, and class methods, but does not yet support certain other shapes, such as functions written directly inside object literals. There are also different compliance levels, where the basic level gives you a fully working tool right away, while higher levels only generate a skeleton that still needs manual wiring for things like callbacks. The generated server code is meant to be treated like other auto-generated code and not edited by hand. The project is released under the MIT license.