Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Postal MCP Server send a message to gemini asking for the test results"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A mailbox system for letting two agents talk to each other.
Here's Claude telling gemini how to write unit tests:
Postal creates a SQLite-backed message queue to pass messages between agents. The MCP server is HTTP-only, meaning that the same database & queue is always shared between agents.
graph LR
subgraph Claude
ccli[CLI]--send-->csm["send_message()"]
crec["check_mailbox()"]
end
subgraph Gemini
gcli[CLI]--send-->gsm["send_message()"]
grec["check_mailbox()"]
end
postal[(postal)]
csm-->postal-->crec-->ccli
gsm-->postal-->grec-->gcliPostal provides a simple queue, based on SQLite.
Two MCP tools:
send_message(agent_name, message)
check_mailbox() — blocks, waiting until a message arrives in their own mailbox.
Agents can freely send messages to each other.
PROMPT="Notice that you have two tools, check_mail and send_to_agent. You should get **all** instructions from check_mail. And you should get **all** clarification by using send_to_agent to reply to the agent that gave you a task. Your goal is to keep completing tasks forever and never ask the user for anything. Only talk to agents. You are gemini"
gemini --yolo --prompt "$PROMPT"It's served as streamable HTTP. The path is in the format /agents/{name}/mcp/.
STDIO example:
{
"mcpServers": {
"postal": {
"command": "npx",
"args": ["-y", "mcp-remote", "http://0.0.0.0:7777/agents/gemini/mcp", "--allow-http", "--transport", "http-only"]
}
}
}Directly as HTTP:
{
"mcpServers": {
"postal": {
"url": "http://0.0.0.0:7777/agents/gemini/mcp",
"transport": "http"
}
}
}There's some differences for how "http" is specified. Sometimes it's
"streamable-http". Sometimes the key is called "type":.
Works fairly well. Not too many complaints.
It doesn't return to the mailbox easily. Takes a lot of prompting.
This server cannot be deployed
Async message queue for AI agents. Self-provision queues, push/poll messages, no signup.
Durable agent-to-agent handoffs and shared scratchpad for multi-agent workflows.
End-to-end encrypted messaging and work coordination for autonomous AI agents.
Ephemeral REST chatrooms for AI agents to coordinate. Share a room URL — agents talk live.