An inbox for Claude that skips the human paperwork

What it does\n@keyid/agent-kit is essentially an MCP adapter for KeyID.ai’s free email infrastructure, giving LLM agents a working inbox without human registration. It exposes 27 JSON-RPC tools—keyid_get_inbox, keyid_send, keyid_reply, thread management, drafts, contacts, and vacation auto-replies—that any MCP client such as Claude Desktop or Cursor can invoke. Agents authenticate with Ed25519 keypairs, which auto-generate on first run, so there is no traditional signup step.\n\nThe interesting bit\nThe project treats the AI agent as the primary user rather than an extension of a human account: KeyID.ai provisions mailboxes to cryptographic identities, not people. That eliminates the usual OAuth dances and API key management, though it also means your agent’s email identity is tied to an external service whose operational details remain largely unexplained.\n\nKey highlights\n- 27 granular tools covering the full email lifecycle, including keyid_forward, keyid_create_draft, and keyid_manage_list for block/allow lists\n- keyid_provision and keyid_get_email let an agent bootstrap its own identity\n- Runs over MCP stdio transport, making it compatible with any MCP-compliant client\n- Free tier with no human-facing signup; Ed25519 keys handle auth automatically\n- Agent-oriented extras like webhook management, forwarding rules, and usage metrics via keyid_get_metrics\n\nCaveats\n- The README is silent on data retention, privacy policy, and rate limits for the free KeyID.ai backend.\n- There is no discussion of deliverability guarantees or what happens if the service changes terms.\n\nVerdict\nUseful for prototyping agent workflows that need to read and write email without wiring up Gmail OAuth. Avoid if you need self-hosted infrastructure or cannot tolerate a third party holding your agent’s mail and keys.