Add the Drin MCP server and four Claude Skills, and Claude — in the desktop app or Claude Code — can send mail, triage an inbox and reply, verify a domain, and report on deliverability without leaving the conversation.
// ~/.claude/mcp.json — illustration; exact setup in the docs { "mcpServers": { "drin": { "command": "npx", "args": ["@drin00/mcp"], "env": { "DRIN_API_KEY": "drin_live_…" } } } } # Then, in a Claude Code session: > Triage my drin inbox and draft replies to anything > from a paying customer.
Register Drin in Claude's mcpServers config with a scoped API key. Claude gains the full 54-tool surface on restart.
Skills teach Claude the right way to use the tools — compose, triage an inbox, verify a domain, and report metrics — so it acts, not guesses.
"Reply to the unanswered threads in my inbox" or "send this receipt" — Claude picks the tools and does the work.
Ask Claude to compose and send transactional email — single messages or batches, with attachments and merge variables.
Claude reads incoming threads as structured events, summarises them, and replies in-thread from the agent's own address.
Hand Claude a domain; it fetches the SPF, DKIM and DMARC records, walks you through DNS, and confirms verification.
Claude reports on delivery, bounce and complaint rates, and flags when a sender is approaching a guardrail.
Claude doesn't just fire off mail — it can read the reply and answer it.
Because Drin was built inbox-first, the same Claude session that sends a message can watch for the response, read the whole thread, and reply in context — from an address that belongs to the agent.
That closes the loop: support that actually resolves, onboarding that follows up, confirmations that handle the "wait, can you change that?" reply. See the agent inbox →
Grab a key, add the MCP server and skills, and Claude can send, triage and reply. Test in the sandbox; production starts at $3 a month.