LLMTEXT (llmtext.com) is an open-source toolkit developed by Jan Wilmake and launched under Parallel_AI's banner in October 2025. It provides three tools to help websites become accessible to AI agents through the llms.txt standard — a format proposed by Jeremy Howard to give LLMs structured, token-efficient access to website documentation.
By late 2025, AI bots had overtaken human visitors as the primary users of the web. Wikipedia reported an 8% decline in human traffic as AI summaries absorbed queries. Yet most websites continued to serve HTML built for human eyes, not structured text optimised for language model consumption.
The llms.txt standard — a Markdown file at the root of a domain, linking to clean text versions of key pages — had been adopted by Anthropic, Cloudflare, Docker, HubSpot, and others, but adoption was uneven and many implementations were technically incorrect.
LLMTEXT addresses both the supply problem (helping websites create valid llms.txt files) and the consumption problem (making valid llms.txt files instantly useful as MCP servers).[1]
Turns any valid llms.txt into a dedicated MCP server installable in one click. Conceptually similar to Context7 but narrowly focused on a single website, and using LLM reasoning — not vector search — to select which documents to load into context.
The server exposes two tools:
The MCP only installs for llms.txt files that fully comply with the spec — a quality gate that incentivises website owners to fix their implementations.
Validates an existing llms.txt. Common errors Jan found across popular implementations:
| Error | Example |
|---|---|
| Oversized table of contents | Cloudflare's llms.txt is 36,000 tokens |
| Wrong content-type on linked pages | Docker, Bitcoin.com serve HTML |
| Non-root hosting | Mintlify, Next.js serve under /docs/ |
| Raw URLs without markdown link format | Cursor |
| Documents too large | Supabase links to an 800,000-token doc |
A framework-agnostic generator powered by Parallel's Extract API. Scrapes a site via its sitemap, converts pages to Markdown, and outputs a spec-compliant llms.txt. Jan used it to generate Parallel AI's own llms.txt.[1]
LLMTEXT extends the Agent_Friendly_Web mission. The mirror network (uithub, openapisearch, etc.) makes existing hostile-to-AI websites accessible by proxying them through clean-text endpoints. LLMTEXT instead helps websites publish clean text proactively, following an open standard — a more scalable and collaborative approach.
It is also Jan's most significant project as a DevRel contractor at Parallel, and the clearest example of his Depth_First_DevRel approach: building a real, widely-useful open-source tool that demonstrates Parallel's capabilities (the Extract API) and drives developer awareness.
Launched October 30, 2025 via a Parallel AI blog post authored by Jan. Adopted by the broader llms.txt community; the MCP server became installable for any valid llms.txt at mcp.llmtext.com.
On December 18, 2025 Jan shipped an update to llmtext-mcp (the GitHub repo behind llmtext.com):
llms-txt-fetch: a library to fetch all contents from a llms.txt fileexplore-sitemaps: a way to quickly get robots.txt, sitemaps, and RSS feeds in a structured formatmap.llmtext.com — a fully stateless sitemap mapper[2]Before the formal LLMTEXT launch, Jan ran an informal quality campaign against broken llms.txt implementations in October 2025. He used an early version of the check tool to audit major sites, finding widespread errors. Notable findings:
Jan's reaction: "It's wild to see how many llms.txt are incorrect or completely unusable! Even big companies..." and "This is terrible for the llms.txt ecosystem!"[3]
He tagged Jeremy Howard (@jeremyphoward, creator of the standard) asking for action, and began correspondence with the affected companies to fix their implementations.[4]
Jan engaged actively in debates about the llms.txt standard throughout November 2025, including:
llms.txt should link to markdown pages, not contain markdown itself — and that llms-full.txt is not part of the spec[6]../blog/2025/LLMTEXT-for-llmstxt.md ↑x-archive/raw/posts/2025-12-18.md ↑x-archive/raw/posts/2025-10-21.md ↑x-archive/raw/posts/2025-10-22.md ↑x-archive/raw/posts/2025-12-14.md ↑x-archive/raw/posts/2026-01-28.md ↑x-archive/raw/posts/2025-11-05.md ↑