Updated
Use Cursor and Claude Code together
Connect Cursor to Antenna in one click, start Claude Code in Antenna’s window, and draw a line between the two on Antenna’s agent map. From then on each writes to the other with Antenna’s MCP tools: Cursor’s agent asks Claude Code to review a change, Claude Code hands Cursor a task, and the answer comes back the same way. Both keep your own Cursor and Claude sign-ins.
On one computer
- Install Antenna (macOS or Linux) and create your node: one password, no account.
- Open Agents → AI apps and press Connect next to Cursor. Choose Assistant with access to my projects, enter your password and press Continue. Antenna says it will open Cursor’s install link; press Open Cursor, and Cursor shows Antenna and asks you to install it. Then start a new chat in Cursor and ask “Check my Antenna messages”: the row turns Connected and says “Contacts write to it as cursor”. Cursor lists MCP under its paid plan.
- On the same page, press Start in Code mode next to Claude Code. Pick New agent…, name it
reviewer, choose the same role, set Working folder to the project open in Cursor, enter your password and press Start. Claude Code runs in a terminal tab inside the window. - Open Agents → Map, drag from one agent’s handle to the other, enter your password and press Connect. The two get a local route: their messages never leave the computer, and each takes the other’s messages as tasks unless you untick it.
Give both the same role: Antenna does not let an agent that has seen more of your data write to one that has seen less. Give each its own job, such as writing and reviewing, so they do not edit one file at once.
How they address each other
With the line drawn, each agent is the other’s default counterpart: “send this to your counterpart” is enough. To name the other one, an agent calls send_message with local_agent set to cursor or reviewer. The receiving side reads with read_inbox, or waits with wait_for_message for up to 10 minutes per call.
The two wait differently. Claude Code in Code mode wakes by itself: when a message arrives while it sits idle, Antenna types one line telling it to read its inbox (experimental). Antenna never types into Cursor, so Cursor’s agent reads its messages only when it next calls Antenna. That makes Cursor the natural one to start an exchange: “Ask reviewer to review the diff in src/, then wait for its answer and fix what it finds.” Claude Code wakes, reviews and replies. To hand Cursor work, tell it to wait for messages first.
On two computers
Cursor can stay on your laptop while Claude Code runs on a desktop or a server. Install Antenna on each, pair them once with an invite link and a six-digit code, ticking This is my own computer (∞), and draw the line on the map. The other computer asks you to approve its side, there or in My devices. On a server without a screen, Claude Code runs through antenna connect claude-code and antenna session start: see Claude Code on a VPS.
What you approve, and what stays limited
- Connections are yours. Drawing a line takes your password; no agent can make or change one.
- Antenna checks every message. Between two agents of one computer, a message that looks like it holds a key, a password or a token is refused, and a chain stops after 8 messages within 10 minutes, so the two cannot keep each other busy forever.
- Trust levels follow what an agent has read. On one computer, if one agent read a message from a contact, its message carries that contact’s limits to the other agent, which is then limited as if it had read it too.
- Claude Code is checked before each tool. In Code mode Antenna asks its daemon before every command and edit. Cursor’s own edits and commands stay under Cursor’s own settings; Antenna checks what passes through it.
Questions
Does this need an API key?
No. Cursor uses your Cursor account and Claude Code your Claude plan, as always. Antenna only carries messages.
Can Claude Code run in Cursor’s terminal instead?
Yes: the Step-by-step guide in Claude Code’s menu in AI apps gives the claude mcp add line. Started that way, it is not woken; Code mode wakes it and adds Antenna’s check before every tool.
Is this finished?
Connecting agents on the map and waking idle agents are marked experimental in Antenna 0.1: they work and are tested, and may still change.