ragrabbit

MCP.Pizza Chef: madarco

The piece that plugs into Claude Desktop or Cursor is small and honest: a single tool that asks your own RagRabbit installation for the most relevant pages, plus one readable summary file. Everything that makes it useful happens beforehand, in a self-hosted web app you deploy yourself with a database and a paid OpenAI account, then point at your site to crawl. Worth knowing: the search route on that installation carries no password at all, so anyone who learns your web address can query your content and spend your credit.

Web/Research

Writing

Use This MCP server To

Ask questions about my own documentation without leaving chat Pull the right help page for a customer question Check what my site already says before writing a reply Give my assistant my product docs as a reference Search my crawled site for a policy I half remember

README

RagRabbit

Self Hosted Site AI Search, LLMs.txt, MCP Server that crawls your content. 1-Click Deploy on Vercel.

RagRabbit

How it works

RagRabbit is a Next.js Turborepo app that uses Llamaindex with pgVector.

Features

💬 Chat Widget: Embeddable AI Chat agent and instant Search
🕸️ Website Crawler: scrapes and index pages with pgVector and PostgreSQL
📄 LLMs.txt Generation: fully customizable wiht ToC reorder
🔌 MCP Server: npx @ragrabbit/mcp to access your docs from Claude Desktop and Cursor IDE
🛠️ Flexible: Authentication, Open Source, API Keys access
🚀 Easy Deployment: One-click setup on Vercel

Integrations:

Fumadocs

ragrabbit FAQ

Can I install this and search any website?

No. It only talks to a RagRabbit installation you have set up and pointed at your own pages. On its own it has nothing to search.

What does it actually cost?

The code is free, but the installation behind it needs somewhere to run, a database, and an OpenAI account with credit, which is charged every time pages are indexed or searched.

How many things can it do?

One. It offers a single document search, plus one plain-text summary of your site it can read. Everything else in the project is the website side, not the assistant side.

Can I use this to answer support questions from my docs?

Yes, and that is the main point of it. Once your pages are crawled, the assistant can pull the closest matching page and quote from it.

Is it safe to leave the installation running?

Treat the address as semi-public. The search route accepts requests from anyone with no password and no request limit, so publishing that address effectively publishes your indexed content.

The project mentions key-based access. Does that protect the search?

Not on this path. The assistant sends no key whatsoever when it searches, so that protection does not apply to the part described here.

How hard is setup?

The snippet you paste into your assistant is trivial. Getting to the point where it returns anything means deploying a whole web app first, which is developer territory.