Vibecoders: Working with AI Coding Assistants

MinT exposes an LLM-readable site-wide documentation summary at https://mint-doc.macaron.im/zh/llms.txt, making it easy for AI coding assistants like Cursor, Claude Code, and Codex to load context in one shot.

Three entry points

/zh/llms.txt: an index of the Chinese pages, one line of summary per page.
/zh/llms-full.txt: the full markdown of all Chinese pages concatenated.
/zh/llms.mdx: the same content as llms-full.txt, provided in MDX form.

Cursor: load with @docs

In Cursor you can use the URL directly as project docs context:

@docs https://mint-doc.macaron.im/zh/llms-full.txt

Claude Code: pull once at the start of a session

Pull the full docs once at the start of a Claude Code session — just once for the whole conversation. Smoke test:

curl -sL https://mint-doc.macaron.im/zh/llms.txt | head -40

Single-page markdown

The site-wide file is a few thousand tokens; you can fetch a single page's content by appending /content.md to the page URL. Example:

curl -sL https://mint-doc.macaron.im/zh/community/customize/sft/content.md

Security tip

When working with AI coding assistants, always scrub API keys starting with sk- from pasted content.

Enterprise differences

Enterprise itself does not need a separate llms.txt — the SDK and docs site come from exactly the same source as Community. A training script that an AI assistant generates from the Community docs runs directly on Enterprise; the only difference is the endpoint and key. You can write the MINT_BASE_URL replacement requirement into your project's CLAUDE.md / .cursorrules so the assistant doesn't forget the config when generating scripts.