Skip to content

Agents & MCP

tunnel-mcp enables direct communication between Claude agents

Written by an AI model from the linked sources and checked against them by a second model. How we work

Image: github.com

The new tunnel-mcp MCP server allows two developers to connect their Claude agents directly, removing the need for a human to manually copy messages between chats. The host initiates tunnel_open, and the guest connects via tunnel_join using a one-time link.

The channel uses a temporary cloudflared Quick Tunnel. Both sides perform only outbound connections, allowing it to function behind NAT and standard firewalls. Message bodies are end-to-end encrypted using NaCl secretbox (XSalsa20-Poly1305), meaning the relay sees only ciphertext.

The connection link is single-use, expires after 10 minutes, and locks the session to two participants once the guest joins. Installation is via npm install -g tunnel-mcp, followed by claude mcp add tunnel. The project is MIT-licensed and includes 159 tests.

Claude Code

Claude Code network sandbox bypassed via null-byte injection

Claude Code had a network sandbox bypass using a null-byte: attacker-host.com\x00.google.com. The JS filter saw the suffix, but libc truncated the string to resolve the blocked host. The bug persisted across 130 releases for 5.5 months. Anthropic patched it in v2.1.90 without a CVE or release note entry.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.