Jupyter MCP Server flaw lets attackers hijack management routes through DNS rebinding

  • Jupyter
  • DNS rebinding
  • security

DNS rebinding flaw affects versions 1.0.2 and earlier

Attackers can terminate kernels and modify server configurations

Version 1.0.3 patches the vulnerability

Jupyter MCP Server maintainers disclosed a DNS rebinding vulnerability in the server, which could let a web attacker reach management routes and change server state. The GitHub advisory says versions 1.0.2 and earlier are affected, and version 1.0.3 patches the flaw.

The server's protected /mcp endpoint rejects forged Host and Origin values, but other custom routes do not. An attacker could use DNS rebinding to reach a server listening locally or on a LAN, VPN, or Docker-published address from a victim's browser.

The flaw could let an attacker stop a notebook kernel through /api/stop, change the configured runtime_url and document_url through /api/connect, and cause the server to send Jupyter API and WebSocket traffic to attacker-controlled infrastructure.

Key terms explained

DNS rebinding

DNS rebinding is a technique where an attacker uses a malicious DNS server to trick a web browser into accessing internal network resources that are not normally reachable from the internet. This allows the browser to act as a proxy for the attacker to interact with internal services.

Think of it this way. It is like a delivery driver who is told to go to a public park, but once they are inside the city, the address is suddenly changed to a private, restricted building.

kernel

A kernel is a separate software process responsible for executing user code and managing computations in an interactive computing environment. It communicates with the user interface to process instructions and return results.