LLM markdown routes should normalize md suffix before source lookup

Problem

Public markdown links need to keep the .md suffix so agents receive a stable URL like /docs/installation/mcp.md.

Fumadocs page lookup should not receive that suffix. source.getPage(['installation', 'mcp.md']) looks for a page named mcp.md, while the real slug is mcp.

Symptoms

• The copy menu needed .md links for ChatGPT and Claude.
• /docs/installation/mcp.md could be rewritten to an LLM route.
• Direct /llm/installation/mcp.md requests missed the page because the route passed the raw slug to source.getPage.
• Chinese markdown routes had the same risk when falling back to English docs.

Solution

Normalize only the route params used for source lookup:

export const stripMarkdownSuffixFromSlug = (slug: string[] = []) => {
  if (slug.length === 0) return slug;

  const last = slug.at(-1);

  if (!last?.endsWith('.md')) return slug;

  return [...slug.slice(0, -1), last.slice(0, -3)];
};

Keep the public URL builder unchanged. The shared URL should stay /docs/... .md; only the internal lookup slug should become /docs/....

Why This Works

There are two contracts:

• Public markdown URLs include .md because that is the user-facing and agent-facing document format.
• Fumadocs source lookup uses normalized content slugs without file extensions.

Keeping those contracts separate makes rewritten docs markdown URLs and direct LLM route URLs behave the same way, including localized /cn routes.

Prevention

Test both entry points:

• /docs/installation/mcp.md
• /llm/installation/mcp.md

For localized docs, test both localized markdown output and English fallback lookup.