Home Getting Started Human Spec LLM Spec

Have an LLM Write Your Spec

Just give an LLM the Author Reference Card and describe your API. It writes the spec—your API documentation—for you.

Think of it like having AI write your OpenAPI document or Swagger file, but in a format that's easier for both humans and AI to understand.

It's just a conversation

Open Claude, ChatGPT, or any LLM. Attach the reference card and describe what you need:

You
JS
Use this reference card to write a MAPI spec for my task management API.
MAPI-AUTHOR.md
You (continued)
JS
Base URL is https://api.taskmaster.io/v1, uses bearer auth.

I need: create task, list tasks, get task by ID, update task, and delete task.

Tasks have title, description, status (pending/done), and due_date.
Claude
C
I'll create a MAPI specification for your task management API following the reference card format...

# Task Management API

[Generates complete spec with all 5 capabilities, TypeScript interfaces, Intentions, etc.]

That's it. The reference card teaches the LLM the format. You just describe your API naturally.

MAPI Author Reference Card

Download this and attach it to your chat
Fetch this page: https://markdownapi.org/specs/MAPI-AUTHOR.md

Or paste this prompt directly into your LLM chat

Have an OpenAPI spec already?

Even easier—just attach both files:

You
JS
Convert this OpenAPI spec to MAPI format using the conversion reference card.
OPENAPI-TO-MAPI.md
openapi.yaml

Iterate naturally

If something's not right, just ask:

  • "The Intentions are too generic—make them explain when to use each capability"
  • "Add a Logic Constraint that due_date must be in the future"
  • "I forgot to mention there's also a bulk delete endpoint"
Tip

The more context you give about your API, the better the spec. Business rules, edge cases, authentication quirks—anything the LLM knows, it can document.

Next steps

Convert from OpenAPI Validate the output