# Introduction (/introduction) Related: [AgentMail](/index.md), [Quickstart](/quickstart.md), [Architecture](/architecture.md) > What AgentMail is, why an agent needs its own inbox, and which surface to start with. # Understand what AgentMail is and where to start AgentMail is an email API built for AI agents. An inbox gives an agent a real address it receives mail at and sends from, both directions driven through the API. Read this page to pick a starting surface, then go to `/quickstart` for the runnable loop. ## Facts * An AgentMail inbox is a real email address. The agent receives mail at the address and sends from the address through the API. * The inbox address appears in the From line of the agent's mail and can live on your own domain (see `/advanced/custom-domains`). * Mail addressed to the agent's inbox stays out of human mailboxes because the address belongs to the agent. * Allow and block lists decide who an inbox can send to and receive from (see `/core/inbound-control`). * AgentMail handles the SPF, DKIM, and DMARC authentication that keeps an inbox's mail deliverable (see `/advanced/deliverability`). * Inbound mail is screened before the agent sees anything. Viruses are rejected at the gateway. Everything else is labeled on arrival (see `/core/receive`). * Labels such as `spam` and `unauthenticated` are hidden from the agent's default view. * When a message lands, a webhook (`/advanced/webhooks`) or WebSocket (`/advanced/websockets`) event wakes the agent. * The agent reads the whole thread for context (`/core/conversations`), then sends its reply through the same API into the same conversation (`/core/send`). * `extracted_text` on a message holds only what the sender just wrote, with quoted history stripped. Attachments such as PDFs, Word documents, and spreadsheets return pre-extracted text via `text_url` (see `/core/receive`). * Errors that reach the API return a stable `code` and a `docs` link, plus a `fix` field stating the corrective action when a generic remedy applies. A taken inbox username returns up to three available alternatives in `suggestions` (see `/advanced/errors`). * Send endpoints accept an `Idempotency-Key`, so a retried send cannot deliver the same email twice (see `/core/send`). * Drafts double as a human-approval gate: the agent writes the draft, a human reviews and sends it (see `/core/send`). * Pods isolate tenants inside an organization, and API keys scope to the organization, one pod, or one inbox (see `/advanced/multi-tenant`). * Receiving email is free. Plans meter outbound sends (see `/advanced/plans-and-usage`). * The Quickstart (`/quickstart`) covers getting a key, creating an inbox, and running the full send, receive, and reply loop from the CLI, curl, or the SDKs. * Integration setup pages exist under `/integrations/` for LangChain, OpenAI Agents SDK, Vercel AI SDK, Google ADK, Mastra, eve, LiveKit Agents, Sim, Replit, OpenClaw, Grok, Hermes, Claude Cowork, OpenAI Codex, Cursor, and Devin. * Any MCP-capable client can connect through MCP and Skills (see `/integrations/mcp-and-skills`). * The four surfaces: CLI and SDKs (`/integrations/sdks-and-cli`), MCP and Skills (`/integrations/mcp-and-skills`), the Console web dashboard at `https://console.agentmail.to`, and the HTTP API reference (`/api-reference/inboxes/list-inboxes`). * Support: the AgentMail Discord at `https://discord.gg/hTYatWYWBc`, or email `support@agentmail.cc` for anything you would rather not post publicly. ## Not supported * Running email flows through AgentMail requires no person watching an email client. * You do not configure SPF, DKIM, or DMARC yourself. AgentMail handles that authentication. * A message carrying a virus is rejected at the gateway and does not reach the agent. * Mail labeled `spam` or `unauthenticated` does not appear in the agent's default view. ## Related * [/quickstart](/quickstart) - create an inbox and run the send-and-receive loop. * [/architecture](/architecture) - the organization, pod, and inbox hierarchy, and where every other resource attaches. * [/core/inbound-control](/core/inbound-control) - allow and block lists for both directions. * [/integrations/mcp-and-skills](/integrations/mcp-and-skills) - connect any MCP-capable client. * [/examples/outbound-gtm-sdr-agent](/examples/outbound-gtm-sdr-agent) - outbound SDR agent that sends, follows up, and reads replies. * [/examples/customer-support-agent](/examples/customer-support-agent) - support agent that answers from docs and past threads. * [/examples/ai-employee-with-its-own-inbox](/examples/ai-employee-with-its-own-inbox) - AI employee teammates can CC and forward to. * [/advanced/multi-tenant](/advanced/multi-tenant) - isolated inboxes, domains, and keys per customer.