August 4, 2026 · 5 min read
Docusaurus is the most widely deployed documentation framework in the JavaScript ecosystem. Meta uses it. Thousands of open-source projects use it. Vercel, Supabase, Prisma, and dozens of other developer tools have shipped Docusaurus-powered docs.
The problem: AI agents can't read Docusaurus sites out of the box. When a user asks Claude or Cursor a question about your product, the AI guesses from training data — which is stale, incomplete, and confidently wrong about your specific configuration options or recent API changes.
This guide fixes that. You'll take any public Docusaurus site and create a live MCP endpoint that any AI agent can query to get cited, accurate answers from the actual documentation.
Docusaurus generates server-rendered HTML — which is great for indexing. But the challenge isn't rendering, it's coverage. A typical Docusaurus site has 50–500 pages across tutorials, API references, guides, changelogs, and version archives. No AI agent can read all of them on demand.
Tools that do live fetching (fetch the page when the AI asks) hit a fundamental wall: the AI doesn't know which page to fetch. It might fetch the homepage and miss the answer buried in the /docs/configuration/advanced page.
The correct solution is pre-indexing: crawl all pages, split them into chunks, generate vector embeddings, and then do semantic search when a question comes in. That way any question — no matter how niche — gets matched against all content, not just whatever page the AI guesses to fetch.
Go to agentready.it.com and paste your documentation root URL. For most Docusaurus sites this is the docs subdirectory:
https://your-project.com/docs/Starting from the docs root ensures the crawler covers all documentation pages. AgentReady follows internal links automatically, so the entire docs tree gets indexed in a single pass — typically 30–90 seconds for a 100-page site.
Docusaurus sites work well with AgentReady because they use server-rendered HTML. Every page delivers its content directly in the initial response, with no JavaScript required to see the text — which means the crawler gets clean, complete content for every page.
After indexing, the fastest way to connect Claude Desktop is with the npx command from the connect page. Open ~/Library/Application Support/Claude/claude_desktop_config.json and add:
{
"mcpServers": {
"agentready": {
"command": "npx",
"args": ["-y", "@agentreadyweb/mcp"]
}
}
}Restart Claude Desktop. You can now ask Claude: "Using the [your-project] docs, how do I configure X?" — and it will search the indexed content and return a cited answer with links to the specific documentation pages.
In Cursor, go to Settings → MCP and add an HTTP server. The endpoint is:
https://www.agentready.it.com/api/mcpThe same endpoint works for Windsurf, Zed, and any other MCP client that supports HTTP transport. Once connected, the AI assistant in your editor can answer questions about your Docusaurus docs mid-workflow — without you switching context to the browser.
If you're the maintainer of the Docusaurus project, indexing it in AgentReady makes it queryable for your users' AI agents too. When a developer asks their AI assistant about your library, the assistant can query the AgentReady index and return a grounded, cited answer — instead of hallucinating about an API that changed two versions ago.
A few setup tips specific to Docusaurus:
Use versioned docs carefully. If your site has version archives (e.g. /docs/2.x/, /docs/3.0/), start from the current docs root to avoid indexing outdated content. AgentReady will follow all links from the starting URL, so starting from /docs/ may include older versions.
Re-index after major releases. Docusaurus makes it easy to ship docs alongside code, but the AgentReady index is a snapshot. Add a re-index step to your release checklist, or use a webhook to trigger a fresh crawl when new content ships.
Host llms.txt at your domain root. After indexing, download the generated llms.txt from your site's detail page and host it at your-project.com/llms.txt. This is the emerging standard for AI agent discoverability — it lets agents find your MCP endpoint without needing to know it exists in advance.
When an AI agent queries the AgentReady index for your Docusaurus site, it receives chunked, source-attributed content from the most relevant pages. The response includes the source URL for each chunk, so the AI can cite exactly where it found the answer.
This matters for developer trust. An answer like "According to your-project.com/docs/configuration#timeout, the default timeout is 30 seconds" is worth far more than a confident guess with no source — especially when the AI is about to be used to write production code.
Several popular Docusaurus-powered sites are already indexed and queryable. Check the AgentReady directory — if your project is listed, developers using Claude or Cursor can already get accurate answers about it. If it's not there yet, indexing it takes under two minutes.