llm-functions

Why we included this project

Most teams already have working code in the shell or in Node or Python, and this project lets them hand that code to an LLM without learning a heavy SDK. You write a plain Bash, JavaScript, or Python function, add a few structured comments, and the tool generates the JSON declarations that make it callable by a model. It is built around OpenAI-style function calling, so the mental model stays small. Beyond single tools, you can assemble agents by combining a prompt, a set of tools, and optional local or remote documents, which is a handy way to prototype something like a custom GPT without vendor lock-in. The included MCP server and bridge mean the tools you build can be consumed by MCP-compatible clients, and external MCP tools can be pulled back in.