Rabbithole MCP
The MCP server connects an agent to a learning canvas where a question about a passage opens a linked document.
- 327 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The server stores local documents, and the agent submits relevant content to its chosen model when answering.
Install
- Works in
Manual install
claude mcp add rabbithole -- npx -y github:shlokkhemani/rabbitholeFirst MCP quick-start command for Claude Code. Start a fresh session after registration.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server stores local documents, and the agent submits relevant content to its chosen model when answering.
This is third-party code. Review the repository files before installing.
What it does
Rabbithole MCP supports learning through linked documents. Users open material on a canvas, ask about a passage and receive an answer in a new document. Branching preserves the path from source text to explanations. In MCP mode, Claude Code, Codex or another connected client answers. The canvas and documents remain on the user’s machine while model requests use the agent’s normal connection. The web app is a separate host.
Who it is for. Students and researchers asking follow-up questions about difficult material while keeping the answer chain visible.
Good fit when
- A difficult term needs a separate explanation while reading.
- One answer opens several research directions.
- A lesson should retain a chain of questions and linked material.
- An already configured coding agent should answer canvas questions.
Not a fit when
- The client cannot extend an overly short tool timeout.
- Documents must not reach the chosen agent’s model.
- You need a shared hosted document store with team accounts.
Example request
Open this document in Rabbithole and create a linked explanation of the selected term.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
MCP requires Node.js 18 or later and a browser; verify both before setup. A tool call remains pending while the agent listens for canvas questions, so a short client timeout can disconnect it. Local storage does not imply local model inference; context reaches your agent’s provider. Web and MCP hosts store documents differently, so check where your work lives and what you export when switching between them.
How to disable. Close the canvas session and remove Rabbithole from the MCP client. Local documents remain in the data directory for separate management.
FAQ
What do I say after connecting?
Start a fresh session and ask to open a document in Rabbithole.
Where are MCP documents stored?
By default under `~/.rabbithole/`, configurable with `RABBITHOLE_DIR`.
Are questions lost on disconnect?
No. Questions are saved so the agent can continue answering them after reconnection.
Do I need a Rabbithole account?
No. The project has no account requirement or hosted document store.
Can I keep a canvas snapshot?
The canvas and frozen snapshots use self-contained HTML that can be stored separately.
Discussion
No questions or reviews yet.
Related
MCP serversThe MCP server lets agents explore prepared documents, sections and related assets while retaining source references.
The MCP server lets agents read and search local Molio vaults, browse file trees and ask questions about their contents.
An MCP server built on Docling: converts PDFs into structured documents, builds new ones and feeds them to RAG
A local document base for your agent: hybrid semantic and keyword search plus a web UI for browsing and uploads
The MCP server answers questions from a personal document collection and returns linked passages with sources.
A self-hosted RAG platform: cited answers over GitHub, Notion, Drive and files, with an MCP mode for Claude Code.
README in short
The web host accepts questions, URLs, Markdown and PDFs. Browser documents are separate from MCP storage.