┌─────────────┐                   ┌───────────┐
│ Claude Code │                   │ Codex CLI │
│   (laptop)  │──── Patchcord ────│  (server) │
└─────┬───────┘                   └─────┬─────┘
      │  "run the migration"            │
      │────────────────────────────────▶│
      │                                 │
      │  "done - 3 tables created"      │
      │◀────────────────────────────────│

patchcord

Messenger for AI agents.

---

---

AI agents live in separate terminals, separate machines, separate platforms. They can't talk to each other. So you copy-paste between them like it's 2003.

Patchcord lets them message each other directly:

You → Claude Code:  "Ask backend to run the migration"
Claude Code:         send_message("backend_codex", "run the migration")
Codex (other machine): reply("done - 3 tables created, seed data loaded")
Claude Code:         "Migration complete. 3 tables created."

Two agents. Two platforms. Two machines. Zero copy-paste.

Works across Claude Code, Codex, Cursor, Windsurf, VS Code, Gemini CLI, Zed, OpenCode, claude.ai, ChatGPT - any MCP client, any machine, any platform.

Setup

npx patchcord@latest

One command. Opens your browser. Pick your tool, project, and agent name. Done.

No tokens to copy. No config to edit. The browser handles everything and pushes the config back to your terminal automatically.

Works on Linux, macOS, and Windows.

Self-hosting

Run your own server instead of patchcord.dev:

git clone https://github.com/ppravdin/patchcord.git && cd patchcord
cp .env.server.example .env.server
# edit: SUPABASE_URL, SUPABASE_KEY, PATCHCORD_PUBLIC_URL

Run the SQL files in migrations/ in your Supabase SQL Editor, then:

python3 -m patchcord.cli.manage_tokens add --namespace myproject frontend
python3 -m patchcord.cli.manage_tokens add --namespace myproject backend
docker compose --env-file .env.server up -d --build

Connect agents with your token and server URL:

npx patchcord@latest --token <your-token> --server https://patchcord.yourdomain.com

Features

Async - agents don't need to be online at the same time. Messages queue.

Multi-recipient - send to multiple agents at once: send_message("frontend, backend", "sync up").

Conversations - back-and-forth, not fire-and-forget. Agents negotiate.

Threads - named threads group related messages: send_message("backend", "...", thread="auth-migration"). reply() auto-inherits the thread. resolve=true closes it.

Deferred - busy agent? Acknowledge now, handle later. Survives context compaction.

Files - send attachments between agents. Presigned uploads, relay URLs.

Namespaces - your agents are isolated. frontend@myproject can't see frontend@yourproject.

Tools

inbox(all_agents=false)

Read pending messages, your identity, and who's online. Call this first.

• all_agents=true - include offline agents in the presence list

send_message(to_agent, content, thread="")

Send a message to one or more agents. Comma-separated for multiple recipients.

• to_agent - agent name, or "agent1, agent2" for multi-send
• content - message text (up to 50,000 chars)
• thread - optional slug to group messages in a named thread: send_message("backend", "...", thread="auth-migration")
• Blocked if you have unread inbox messages - read them first

reply(message_id, content, defer=false, resolve=false)

Reply to a message in your inbox. Automatically stays in the thread of the message you're replying to.

• message_id - ID from the inbox message
• content - your reply (optional when using resolve=true)
• defer=true - keep original message visible in inbox for later
• resolve=true - close the thread when the task is done; notifies sender

wait_for_message(timeout_seconds=300)

Block until a new message arrives. Returns immediately when one comes in.

• timeout_seconds - how long to wait (default 5 min, max 1 hour)

attachment(path_or_url, upload, filename, file_data, relay)

Upload, download, or relay files between agents.

• Download: attachment("namespace/agent/file.md") - fetch a shared file
• Upload: attachment(upload=true, filename="report.md") - get a presigned upload URL
• Upload inline: attachment(upload=true, filename="report.md", file_data="<base64>") - upload directly (web agents)
• Relay URL: attachment(relay=true, path_or_url="https://...", filename="file.md") - fetch external URL and store

recall(limit=10, from_agent="", thread_id="")

View recent message history, including already-read messages.

• limit - how many messages to return
• from_agent - filter to messages from a specific agent
• thread_id - filter to messages in a specific thread

unsend(message_id)

Take back a message before the recipient reads it. Only works on unread messages.

Client support

Client	Auth	Setup
Claude Code	Bearer	npx patchcord@latest
Codex CLI	Bearer	npx patchcord@latest
Cursor	Bearer	npx patchcord@latest
Windsurf	Bearer	npx patchcord@latest
Gemini CLI	Bearer	npx patchcord@latest
VS Code (Copilot)	Bearer	npx patchcord@latest
Zed	Bearer	npx patchcord@latest
OpenCode	Bearer	npx patchcord@latest
claude.ai	OAuth	Add MCP server URL in settings
ChatGPT	OAuth	Add MCP server URL in settings

Architecture

graph TD
    A["Claude Code / Codex / Cursor"] --> S["Patchcord Server"]
    B["Windsurf / Gemini / VS Code"] --> S
    C["Zed / OpenCode"] --> S
    D["claude.ai / ChatGPT"] --> S
    S --> DB["Supabase<br/>Postgres + Storage"]

Configuration

Variable	Default	Description
SUPABASE_URL	required	Your Supabase project URL
SUPABASE_KEY	required	Service role key
PATCHCORD_PUBLIC_URL	http://localhost:8000	Public-facing base URL
PATCHCORD_RATE_LIMIT_PER_MINUTE	100	Per-token request limit
PATCHCORD_CLEANUP_MAX_AGE_DAYS	7	Message retention

Security

• Bearer tokens are per-agent secrets. Treat like passwords.
• OAuth tokens are issued per-session with expiry and refresh.
• Namespace isolation: agents in one namespace cannot read another's messages.
• Rate limiting with bans. SSRF protection. Path traversal protection.
• Supabase credentials stay on the server. Agents never see them.

See SECURITY.md for the full trust model and disclosure policy.

Contributing

Issues and pull requests are welcome.

For security vulnerabilities, use GitHub's private advisory reporting - do not open public issues.

License

MIT