fix: md to pdf

[tomaszantas]

Description

Fix api/md-to-pdf.

[vercel]

@tomaszantas is attempting to deploy a commit to the langfuse Team on Vercel.

A member of the Team first needs to authorize it.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[felixkrrr]

@claude pls review

[felixkrrr]

@jannikmaierhoefer can u pls approve the dev deployment?

[felixkrrr]

@claude pls review

[felixkrrr]

Thanks for looking into this — the md-to-pdf endpoint is definitely broken in production and the Chromium bundling fixes are solid work. A few things need to be addressed before we can merge.

The core issue

langfusePathToMdSrcPath() patches one consumer (the md-to-pdf API route), but the underlying problem affects all .md URL access — the "Copy as Markdown" button, direct .md URLs, and content negotiation are all broken for the same pages.

We tested production and found 30 broken .md URLs:

23 marketing pages (all except /support.md which has a one-off beforeFiles rewrite):
/about.md, /careers.md, /cn.md, /community.md, /cookie-policy.md, /enterprise.md, /find-us.md, /imprint.md, /jp.md, /jp-cloud.md, /kr.md, /non-profit.md, /oss-friends.md, /press.md, /pricing.md, /pricing-self-host.md, /privacy.md, /research.md, /startups.md, /talk-to-us.md, /terms.md, /watch-demo.md, /wrapped.md

7 user story pages (URL path is /users/* but md-src/ has them under customers/):
/users.md, /users/canva.md, /users/cresta.md, /users/khan-academy.md, /users/magic-patterns-ai-design-tools.md, /users/merckgroup.md, /users/sumup.md

The root cause is in scripts/copy_md_sources.js — it copies files preserving the content/ directory structure, but the URL structure differs for these sections:

Content path	md-src/ path (current)	URL path	Result
content/marketing/terms.mdx	md-src/marketing/terms.md	/terms.md → rewrite → md-src/terms.md	404
content/customers/canva.mdx	md-src/customers/canva.md	/users/canva.md → rewrite → md-src/users/canva.md	404

Suggested fix

Fix copy_md_sources.js to copy files to paths matching the URL structure, not the content directory:

• content/marketing/terms.mdx → md-src/terms.md
• content/customers/canva.mdx → md-src/users/canva.md

The mapping from content directory → URL path already exists in lib/source.ts (marketing baseUrl: "", customers baseUrl: "/users"), so the copy script should use the same source of truth.

This way the existing generic rewrite /:path*.md → /md-src/:path*.md works for all consumers without any remapping code. langfusePathToMdSrcPath(), lib/marketing-slugs.ts, and the one-off /support.md beforeFiles rewrite all become unnecessary.

What to keep

• outputFileTracingIncludes for Chromium binaries in next.config.mjs — this is correct and necessary.
• Improved error handling/logging for chromium.executablePath() and the SPARTICUZ_CHROMIUM_BIN_DIR override — solid additions.

What to change/remove

• tsconfig.tsbuildinfo — remove from the commit, it's a 1.4MB build artifact.
• langfusePathToMdSrcPath() and lib/marketing-slugs.ts — should become unnecessary once the copy script is fixed.
• The /support.md beforeFiles rewrite already on main — can be removed as part of this fix.

[felixkrrr]

Correction to my review above: The "Copy as Markdown" button only exists on docs-style pages (docs, self-hosting, integrations, guides, library) — not on marketing pages or user stories. So the broken .md URLs for marketing/user-story pages don't affect that button.

The practical impact of the 30 broken .md URLs is:

• md-to-pdf — the "download as PDF" links on /terms and /privacy (which this PR targets)
• Direct .md access — anyone (or any LLM/tool) appending .md to a marketing or user-story URL
• MCP getLangfuseDocsPage — if it fetches .md URLs for these pages

The rest of the review stands as-is.

[felixkrrr]

To clarify the above — here's why fixing this inside the API route isn't the right layer.

The .md URLs are broken for 30 pages in production (23 marketing + 7 user stories). The root cause is a mismatch in scripts/copy_md_sources.js: it copies files preserving the content/ directory structure (content/marketing/terms.mdx → md-src/marketing/terms.md), but the URL structure serves them at a different path (/terms). So the generic rewrite /:path*.md → /md-src/:path*.md fails.

This PR adds langfusePathToMdSrcPath() to remap paths inside the md-to-pdf API route. The problem with that approach:

1. It only fixes one consumer. The same .md URLs are used by direct access (anyone appending .md), the MCP server's getLangfuseDocsPage tool, and content negotiation. Each would need its own copy of the remapping logic.
2. It introduces a hardcoded MARKETING_SLUGS list that has to stay in sync with content/marketing/ manually — a maintenance trap. Every new marketing page would silently break until someone remembers to update the list.
3. It duplicates routing knowledge that already exists in lib/source.ts (marketing baseUrl: "", customers baseUrl: "/users"). The copy script should use the same source of truth instead of re-deriving it.

If we fix copy_md_sources.js to output files at URL-matching paths (content/marketing/terms.mdx → md-src/terms.md, content/customers/canva.mdx → md-src/users/canva.md), the generic rewrite handles everything — no remapping code, no slug list, and all consumers work automatically.

(Correction to my earlier review: the "Copy as Markdown" button only exists on docs-style pages, not marketing/user-story pages, so it's not directly affected here.)

used suggest changes falsely