agent-mail

Agent coordination systems prevent edit conflicts by providing memoizable identities, inbox/outbox messaging, and advisory file reservations (leases) that signal intent. They use a Git+SQLite-backed audit trail to manage threaded discussions and avoid duplicate work across agents.

Git and SQLite backing works by storing searchable communication history in Git commits while indexing it in SQLite. This provides a disaster recovery mechanism and enables static exports for offline or archived review of the audit trail.

Yes, MCP messaging applies across both single-repo and multi-repo setups, enabling per-agent inboxes and threaded discussions. It coordinates multiple coding agents by providing memoizable identities bound to each agent's program and model.

Start the MCP server using a provided startup script or alias, then access the interface at a local port to view inboxes. The Human Overseer workflow integrates with advisory file reservations and a pre-commit guard for priority management.

Advisory file reservations (leases) are the best way to manage concurrent edits, signaling intent to avoid conflicts. A pre-commit guard enforces these leases, while a Human Overseer workflow allows priority changes when reservation conflicts arise.

Yes, FastMCP HTTP transport and Python tooling are required dependencies for running agent coordination. The system relies on these components to provide the MCP server, memoizable identities, and the disciplined data-audit approach needed for operation.