telepathy MCP server
Commit Nov 9, 2026Enables coding agents like Claude Code, Codex, and OpenCode to discover, message, and collaborate with each other across devices.
Also known as Winterrks/telepathy
- Stars
- 5
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 9, 2026
- License
- MIT
What is telepathy MCP?
telepathy enables coding agents like Claude Code, Codex, and OpenCode to discover, message, and collaborate with each other across devices.
Summary generated from the project's README
About telepathy MCP server
Coding agents can communicate, share tasks, and coordinate work across different devices using telepathy. This system allows tools like Claude Code, Codex, and OpenCode to discover one another and exchange messages without manual intervention. It works on a single machine or across a network of devices using a bridge via Tailscale or local connections. Users can pair new devices using a 6-digit code to create a unified network of agents.
Developers use this to prevent agents from colliding on the same files or to delegate specific tasks to specialized assistants. For example, a session on a laptop can trigger a Codex agent on a desktop to run GPU tests or handle a code review in a background worktree. The setup is managed through a command line interface that automates installation across all detected agents. It supports 11 different agents including Cursor, Devin, and Copilot CLI, ensuring that idle sessions are woken up when a new message arrives.
What you can do with it
- 1Coordinate multi-agent workflowsAsk one agent to delegate a specific task like a code review to another agent.
- 2Prevent file edit collisionsHave agents check if other sessions are currently working on the same repository to avoid conflicts.
- 3Execute remote hardware tasksTrigger an agent on a different machine to utilize its specific hardware or files.
- 4Automate cross-device status checksQuery an agent running on a desktop from a session active on a laptop.
Written from the project's README. Every use case is checked against a quote from it.
Automated checks
Ran Nov 15, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablewinterrks/telepathy is public and reachablegithub.com/winterrks/telepathy ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-09github.com/winterrks/telepathy ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubYour agents, working together
You have Claude Code open in one window and Codex in another, maybe OpenCode on your desktop. They work on the same code and don't know the others exist, so you end up copying messages between windows.
With telepathy, every session can see the others, message them and hand them work. Two agents on the same repo notice each other, agree on who changes what, and from then on ask each other for what they need. You don't relay a thing.
Get started
npm install -g telepathy-cli && telepathy setup
telepathy setup finds every coding agent on your machine and installs telepathy in each one. Restart the sessions
you already had open, then ask one agent about another:
Ask the Codex session what it's working on.
| Agent | Install |
|---|---|
| Claude Code | claude plugin marketplace add Winterrks/telepathy && claude plugin install telepathy@telepathy |
| Codex | codex plugin marketplace add Winterrks/telepathy && codex plugin add telepathy@telepathy, then approve its hooks in /hooks |
| OpenCode | opencode plugin add telepathy@git+https://github.com/Winterrks/telepathy.git |
| Kilo Code | add "git:github.com/Winterrks/telepathy" to plugin in ~/.config/kilo/kilo.json |
| Copilot CLI | copilot plugin marketplace add Winterrks/telepathy && copilot plugin install telepathy@telepathy |
| Qwen Code | qwen extensions install https://github.com/Winterrks/telepathy/archive/refs/heads/main.tar.gz --auto-update |
| Grok | grok plugin install Winterrks/telepathy#plugin --trust |
| Devin | devin plugins install Winterrks/telepathy#plugin |
| Antigravity | agy plugin install https://github.com/Winterrks/telepathy/tree/main/antigravity |
| Cursor | cursor-agent plugin marketplace add https://github.com/Winterrks/telepathy, then install it from /plugin |
| Kimi Code | /plugins install https://github.com/Winterrks/telepathy in a Kimi Code session |
You need Node.js 20 or newer. Each agent's page has its update command and its quirks.
What it does
Agents message each other
Any session can reach any other: Claude Code to Codex, Codex to OpenCode, Copilot to Claude. The message lands in
the other agent's conversation, clearly labeled as coming from another agent, and in Claude Code, Codex, OpenCode
and others it wakes an idle session up on its own. In Claude Code, the built-in ListAgents and SendMessage reach
the other agents too.
Tell the Codex session working on the API that the schema migration landed.
Delivery you can count on
A message counts as sent only once it is in the other agent's conversation. If it can't get there, the sender is told so and the message is withdrawn, so it never turns up later as a surprise.
They stay out of each other's way
When an agent starts on files that another running session has just been editing, telepathy tells it which session that is and suggests asking first. Two agents in one repo split the work instead of colliding.
Is anyone else working in this repo? If so, tell them which files you're about to change.
Hand work to any agent
One agent can start another as its helper: Claude Code can start Codex for a review, Codex can start OpenCode to write the tests. The helper works in the background, in its own git worktree if you like, and its report comes back by itself. In Claude Code it is a background task, with no model in between, that you can follow and stop like Claude's own.
Have Codex review the diff on this branch in a worktree while you keep going on the UI.
Across all your devices
Run the bridge on each device and the sessions on your laptop and your desktop join one network, over Tailscale or your local network. Devices on your own tailnet trust each other right away. Anything else pairs once with a 6-digit code that both screens show.
Ask the Codex on my desktop whether the training run finished.
Start agents on your other machines
New in 0.13: start an agent on another of your devices, where it runs with that machine's files, sign-ins and hardware. Its report comes back to the session that asked. Only devices you allow can start one.
Run the GPU tests on my desktop with Codex.
Any agent, instantly
Some agents only pick messages up when their next turn starts. Start one with telepathy run <agent> and every
message reaches it the moment it is sent, even while it is idle.
Grok Bot joins from the cloud
Connect a Grok Bot routine with telepathy grokbot connect and the Bot becomes a session like any other: a message
starts a run with the message in hand, and the Bot can answer.
Set up once, stays current
telepathy setup installs telepathy in every agent it finds and approves its hooks. After that, every copy updates
itself, and your devices move to a new version together.
Private by default
No telemetry, and no server of ours in the middle. Messages stay on your machine unless you start the bridge or connect a Grok Bot, and the bridge only talks to your own devices. What leaves your machine, and when.
Works with
| Agent | A new message |
|---|---|
| Claude Code | wakes it up |
| Codex | wakes it up, or reaches it after its next tool call when it's busy |
| OpenCode | wakes it up |
| Kilo Code | wakes it up |
| Grok | wakes Grok Build up, or starts a Grok Bot routine run |
| Antigravity | wakes the App up; the CLI sees it at its next turn |
| Copilot CLI | arrives at its next turn |
| Cursor | arrives after its next tool call |
| Devin | arrives at its next turn |
| Qwen Code | arrives at its next turn |
| Kimi Code | arrives at its next turn |
Under telepathy run, every agent with an ACP mode gets messages the moment they arrive.
Which versions are tested, and what isn't supported yet.
Learn more
How telepathy behaves, one topic per page, in docs/behaviour:
- How it works: the tools every agent gets, and what runs underneath
- Sessions: addresses, names, statuses, and resuming a session
- Messages: when a message counts as delivered, and what the receiver sees
- Agents: each agent's install, update and quirks
- The telepathy command:
setup,update,runand the rest - Devices: the bridge, pairing, and devices that can't accept connections
- Subagents:
start_agent, on this machine or another - Privacy and security
- Configuration and troubleshooting
- Development
telepathy is young, and some agents have quirks. An interrupted Codex, for one, waits for your next prompt before it picks up messages. If you find one, open an issue.
License
MIT © Winterrks
telepathy MCP questions
How do I get started with telepathy?
You can install it globally using npm and run the setup command to find and install it in every coding agent on your machine.
Can agents work together on different machines?
Yes, by running the bridge on each device, sessions on different machines can join one network via Tailscale or your local network.
Does telepathy send my data to a central server?
No, it is private by default with no telemetry and no central server; messages stay on your machine unless you use the bridge or connect a Grok Bot.
Can one agent start another agent to help with a task?
Yes, an agent can start another as a helper, such as Claude Code starting Codex for a review, and the report will return automatically.