Intercom

Claude Intercom

Two-way communication bridge between Claude Code sessions using the Channels API.

Let two Claude Code instances on different machines talk to each other in real-time. One sends a message, the other receives it instantly as a channel notification and can reply back.

What Intercom does that native can't

Claude Code has native cross-session messaging now. Here's where Intercom is still the answer:

• Two different people. This is the big one. Native connects *your* sessions — the inbox is tied to your own OS user and cross-machine delivery rides your own Remote Control. A backend dev on their laptop and a frontend dev on theirs are two accounts, and native can't join them. That two-person hotline is exactly what Intercom was built for.
• Native Windows. Native messaging is macOS and Linux only. Intercom is Bun over HTTP and doesn't care what OS you run.
• Your network, not Anthropic's. Native cross-machine messages relay through Anthropic servers. Intercom POSTs straight over your own network, VPN or tunnel.
• Any provider. Works on Bedrock, Vertex and Foundry, and in telemetry-disabled environments, where native messaging is switched off.
• Container to host. Native discovery works through files on disk, so a session in a container and one on the host can't see each other. Intercom just needs a reachable port.
• Honest delivery. send_message reports sent, delivered-to-process and answered as three separate states, plus a check_message tool. You find out when the other side never picked it up, instead of planning around an answer that was never coming.

If you're on macOS or Linux and you just want your own sessions talking to each other, native is simpler — use that. Everything else above is what this is for.

Why?

If you have a backend developer and a frontend developer each running Claude Code on separate machines, they currently have to relay questions through Slack/Discord/copy-paste. Claude Intercom creates a direct hotline between the two AI sessions — one Claude can ask the other about endpoints, schemas, or implementation details and get answers from the actual codebase.

Architecture

Machine A                          Machine B

(e.g. backend dev) (e.g. frontend dev)

Claude Code A Claude Code B | | +-- intercom.ts ---HTTP POST----> intercom.ts --+ | (channel) <--HTTP POST-- (channel) | | | +-- stdio (MCP) --+ +-- stdio (MCP) -----+ | | Claude A Claude B

Both instances run the same intercom.ts file. Each listens for HTTP messages and pushes them into its local Claude Code session via the Channels API. Each also exposes send_message and check_message tools that Claude can call.

Requirements

• Claude Code v2.1.80 or later
• Bun runtime
• Both machines must be able to reach each other over HTTP. Tailscale is the recommended way — see Connecting the two machines

Set it up with Claude

Paste this into a Claude Code session on each machine and it will walk you through the whole thing. Run it on machine A first, keep the secret it gives you, then run it on machine B.

Set up claude-intercom on this machine so this Claude Code session can message

a Claude Code session on my other machine.

Work through this in order, and stop and ask me whenever you need something only I can tell you.

1. Check prerequisites. Confirm bun --version works and that claude --version is 2.1.80 or newer. If Bun is missing, tell me how to install it for my OS and stop there.

1. Work out how the two machines will reach each other. Run tailscale ip -4. If that returns an address we'll use Tailscale. If Tailscale isn't installed, tell me it's the recommended option and ask whether I want to install it or use something else.

1. Ask me these, one at a time:

• a short role name for THIS machine, e.g. "frontend", "vps", "laptop"
• the address of the OTHER machine (hostname or IP)
• whether this is the first machine I'm setting up or the second

1. Handle the shared secret.

• First machine: generate a strong random secret, show it to me once, and tell me I'll need it when I run this on the other machine.
• Second machine: ask me to paste the secret from the first one. Both machines must end up with exactly the same secret.

1. Write .mcp.json in this project with an "intercom" server: command npx, args ["-y", "claude-intercom"], and env MY_ROLE, REMOTE_HOST (the other machine plus :8788), INTERCOM_SECRET, INTERCOM_PORT set to 8788. If .mcp.json already exists, merge the intercom entry into it instead of overwriting the file. Make sure .mcp.json is gitignored, it holds the secret.

1. Tell me to restart Claude Code on this machine with: claude --dangerously-load-development-channels server:intercom and explain that the flag is needed because the channels capability is still experimental, and that messages only arrive while both sides are running.

1. Once both machines are up, tell me to test it by asking one session to send a message to the other. Remind me that send_message reports sent / delivered-to-process / answered separately, so "delivered" means the remote process accepted it, not that the other Claude has read it.

Don't invent configuration keys. The only environment variables are MY_ROLE, REMOTE_HOST, INTERCOM_SECRET, INTERCOM_PORT and INTERCOM_SEND_TIMEOUT_MS.

Prefer to do it by hand? The manual steps are below.

Quick Start

1. Install

Nothing to install. npx fetches it on first run, on both machines.

You do need Bun on your PATH — the server uses Bun.serve for its HTTP listener, and the npx entry point hands off to it.

Prefer to run from source?

git clone https://github.com/MuhammadTalhaMT/claude-intercom.git

cd claude-intercom bun install

Then use "command": "bun" with "args": ["/path/to/claude-intercom/intercom.ts"] in the config below instead of the npx form.

2. Configure

Copy the example config into your project's .mcp.json.

First generate a secret, and use the same value on both machines:

openssl rand -base64 32

Then export it in the shell you launch Claude Code from, rather than typing it into the config:

export INTERCOM_SECRET='the-value-you-just-generated'

.mcp.json is a project file, and project files get committed. Claude Code expands ${VAR} in .mcp.json — in command, args, env, url and headers — so the config below can be checked in and shared with your teammate while the secret itself never leaves your environment.

Machine A (e.g. backend — static IP or VPS):

{

"mcpServers": { "intercom": { "command": "npx", "args": ["-y", "claude-intercom"], "env": { "MY_ROLE": "backend", "REMOTE_HOST": "MACHINE_B_IP:8788", "INTERCOM_SECRET": "${INTERCOM_SECRET}", "INTERCOM_PORT": "8788" } } } }

Machine B (e.g. frontend — can be behind NAT):

{

"mcpServers": { "intercom": { "command": "npx", "args": ["-y", "claude-intercom"], "env": { "MY_ROLE": "frontend", "REMOTE_HOST": "MACHINE_A_IP:8788", "INTERCOM_SECRET": "${INTERCOM_SECRET}", "INTERCOM_PORT": "8788" } } } }

3. Start

On both machines:

claude --dangerously-load-development-channels server:intercom

4. Talk

In either Claude Code session:

"Send a message to the other developer asking what API endpoints are available for the dashboard."

Claude will use the send_message tool to POST the message to the other machine. The other Claude receives it as a channel notification and responds.

Connecting the two machines

Use [Tailscale](https://tailscale.com). Install it on both machines and they get stable private addresses on your own tailnet. Then point each side at the other's tailnet address:

"REMOTE_HOST": "other-machine:8788"

This is strictly better than exposing a port to the internet: no public listener, no port forwarding, the address doesn't change when your ISP reassigns your IP, and device identity is enforced by Tailscale rather than resting entirely on a shared string. Set INTERCOM_HOST to your tailnet address — or to 127.0.0.1 if you are also fronting it with a tunnel — to be certain nothing outside can reach it at all.

Alternative: ngrok

If you can't use Tailscale, ngrok still works:

# On the machine behind NAT

ngrok http 8788

"REMOTE_HOST": "your-subdomain.ngrok-free.app"

ngrok hostnames are detected and switched to HTTPS automatically. For any other tunnel — Cloudflare, Caddy, a reverse proxy of your own — write the scheme into REMOTE_HOST explicitly (https://your-host), or the secret goes out over cleartext HTTP.

Note this does put a publicly reachable endpoint in front of your Claude session, gated only by the shared secret — pick a strong one, and consider INTERCOM_HOST=127.0.0.1 so only the tunnel can reach the listener.

Tools

send_message

Sends a message and returns immediately with an id. It does not wait for an answer — the reply arrives later as its own inbound message.

The acknowledgement deliberately does not claim the other developer received it. An HTTP 200 proves the remote *process* accepted the message, not that the other Claude ever read it.

check_message

Answers the question a fire-and-forget channel otherwise can't: is this reply slow, or is it never coming? Pass an id, or omit it to list everything outstanding.

Every message sits in one of three states:

expectReplyWithin is what makes "overdue" mean anything, and it's per message rather than one global timeout — so a quick endpoint lookup and a long investigation aren't judged on the same clock.

Configuration

| INTERCOM_HOST | No | 0.0.0.0 | Interface to b