MCP server

AI agents read the inbox and propose drafts — never auto-send.

What is MCP?

The Model Context Protocol (MCP) is an open standard through which AI assistants talk to external systems. Astreo's MCP server makes the inbox accessible to agents — so an assistant can read context and propose drafts, without ever sending automatically.

Connecting

Connect an MCP-capable client (e.g. an AI assistant) to Astreo's MCP server and authenticate with a personal access token. The agent gets the same scoped view as the associated user — never more.

Available tools

The server provides tools, including:

• Read threads — fetch unread/assigned conversations and read the history.
• Draft a reply — e.g. draft_reply({ thread_id }) creates a draft that a human reviews and sends.

The exact tool list is announced by the server on connect (discovery).

Safety & limits

• No auto-send: AI outputs are drafts; sending stays a human decision.
• Scoped: the agent only sees what the token allows.
• Revocable: block the token any time in the settings.