AgentSync

AgentSync

🔗 AnyDesk for Claude Code sessions

Connect any two Claude Code sessions — on one machine or across the internet — and let them ask each other questions.
A consent handshake to pair, end-to-end encryption on the wire, and a human in the loop the whole time: accept, pause, resume, or stop on demand.

Quickstart · Features · How it works · Tools · Security · FAQ · Roadmap

AgentSync is an open-source Claude Code plugin (an MCP server) that turns isolated AI coding sessions into a collaborating mesh. Inspired by AnyDesk and TeamViewer, it lets one Claude Code session request a connection to another by a short ID, the other side accepts a consent prompt, and from then on the two agents can ask each other questions, exchange messages, and pass results back and forth — bidirectionally, with a human able to intervene at any moment.

It works the same whether the sessions sit in two terminals on your laptop or on two machines on opposite sides of a corporate firewall, NAT, or VPN — because, like AnyDesk, both ends only dial outbound to a rendezvous relay.

[!TIP] New in v0.1.1 — autonomous answering. A session can now answer peer questions on its own by running a locked-down, read-only claude -p — no human needed on the other side. Toggle it live with /agentsync-auto on. Jump to it ↓

💡 Why AgentSync?

A single Claude Code session only knows its own repository. Real engineering, though, spans many repos, machines, and contexts at once — and copy-pasting between terminals to keep agents in sync is slow, lossy, and error-prone.

AgentSync lets specialized sessions cooperate directly: