Skip to main content
A coding agent writes better code for an API when it reads that API’s docs rather than recalling them, and the haau3 platform API may be newer than its training data. docs.haau3.com gives an agent four ways in.

The docs as Markdown

The haau3 skill

docs.haau3.com/skill.md says what the platform does and how to call it, in the Agent Skills format. To add it to your project’s coding agents:
It asks which agents to add the skill to, and puts it in each one’s skills folder, such as .claude/skills/haau3 for Claude Code. Add -g to add it for every project instead.

The docs’ MCP server

https://docs.haau3.com/mcp is an MCP server whose tools search the docs and read any page. It needs no key. To add it:
For Cursor, the entry goes in .cursor/mcp.json in your project, or ~/.cursor/mcp.json for every project. Each page’s menu also has Connect to Cursor and Connect to VS Code.

An index for your AGENTS.md

Paste the block for your SDK into your project’s AGENTS.md, or CLAUDE.md, which Claude Code reads. It names every method with its request and its page as Markdown, in under 8 KB. In Vercel’s evaluation of coding agents on APIs their training data did not hold, an index of about that size in AGENTS.md raised the pass rate from 53% to 100%.
Each block is generated from its SDK’s own api.md, so a new release of the SDK brings a new block.

The SDKs’ own docs

Every method of each SDK carries the API’s own description of its endpoint, in a TypeScript doc comment, a Python docstring or a Go doc comment, so an agent working in your editor reads it beside the call. go doc github.com/haau3/haau3-go prints the Go package’s documentation.