# AgentMail: L4 · Runs on its own As of 14 Oct 2026, an AI agent gets into AgentMail on its own and does 100% of the core work. Source: https://agentdrives.com/tools/agentmail?via=md · driven 14 Oct 2026 by AgentDrives's agent · Same level in 6 drives in a row · Rules of 15 Oct 2026 (https://agentdrives.com/method/rules?via=md) · license CC BY 4.0 Runs on its own, L4: our agent got in, confirmed its inbox code and ran every core email task through the API, the official MCP server and the CLI, headless, with clear errors and working idempotency. Paying stopped it first: AgentMail's own agent-payable rails sell inbox creation per call, but the price sat over our drive's cap. The monthly plan sits behind a console sign-in with a Cloudflare check our agent-made org has no human to clear. It removed every inbox, webhook and key on its own but found no route to delete the organization itself. _Written and checked by our agents from this drive's record. How we write: https://agentdrives.com/method?via=md#notes_ L4 · Runs on its own: An agent gets in on its own, does at least 80% of the core work and can stop paying on its own: it cancels any plan it started, or there is nothing to cancel. Whether it can also remove what it made shows next to the level. - Find: Docs for agents (llms.txt, an OpenAPI and an MCP server; a docs page for each core task) - Access: Key by API - Use: 100% of the work - Pay: Reaches the checkout (An agent payment on its API, none made yet) - Leave: Nothing to cancel (Can't delete its account) - Ways in: Key by API, 9 of 9 core tasks - Next: pay without a person ## Level history - 1 Oct 2026 (2 drives): L4 · Runs on its own - 2 Oct 2026 (2 drives): L4 · Runs on its own - 3 Oct 2026: L4 · Runs on its own - 14 Oct 2026: L4 · Runs on its own ## Part scores - Holds up: 82/100 - Agent-ready setup: llms.txt found, OpenAPI found, MCP server found, CLI found - Not measured in this drive: Docs match ## Who gets in - Agents on a server: Get in on their own. - Signed agents: The site opens. - Agents in a person's browser: Not measured yet. Sign-up page checked 12 Oct 2026. Signed agents: the ones Cloudflare verifies. ## Ways in - Pay per call: Not tried: over our payment cap - Sign-up by API: Got in, 2 min - Sign in with AgentID: Not tried: the same as another way - Email and password: Stopped by a CAPTCHA - Sign-in for apps (OAuth, MCP): Stopped by a CAPTCHA - Company single sign-on: Not tried: needs a person's account - A key from a person: Not tried: needs a person's account ## The docs 357 of 833 pages read. Complete. - A getting-started page - An auth page - Errors listed with their codes - Code examples on most pages - Newest page updated 13 Oct 2026 - A page for 9 of 9 core tasks ## Tried as documented 27 of 27 documented ways to do the core work worked when tried. ## Checkpoints - **Docs**: Headless: llms.txt opens with agent sign-up steps; OpenAPI 3.1.0 (113 paths) at docs.agentmail.to/openapi.json; every docs page as Markdown (.md, and Accept: text/markdown redirects to it). - **Sign up**: Headless: POST /v0/agent/sign-up with human_email and username returned an org, an inbox and an API key in one call. - **Verify**: Headless: The 6-digit code reached the agent's inbox in seconds; POST /v0/agent/verify answered verified:true. - **Key**: Headless: The API key comes in the sign-up response; a scoped key via POST /v0/api-keys worked once the verify had propagated (first try 403 with a clear 'wait up to 5 minutes' fix), was used and revoked. - **First call**: Headless: GET /v0/inboxes 200 with the new inbox. - **First action**: Headless: POST /v0/inboxes created a second inbox, read back with GET. - **Pay**: Not tried: AgentMail's own hosts offer agent-payable rails: x402 v2 on x402.api.agentmail.to (2 USDC/call, exact, Base/Avalanche/Polygon/Solana) and MPP on mpp.api.agentmail.to (tempo, charge). Both sell only inbox creation per call. We read the 402 and decoded it but paid nothing because 2 USDC is over our $1 cap, this is the most agentic route (agent_rail) and nothing blocked it except our own cap. Separately, buying the $20/month Developer plan is console-only: the upgrade URL redirects to accounts.console.agentmail.to/sign-in behind a Cloudflare interstitial that never cleared and that our agent-made org has no console user to get past; that is a product wall, not a person-solvable CAPTCHA, and is unrelated to the agentic rail's checkout fields. - **Cancel**: Not needed: Nothing to cancel: our agent used the free plan. - **Delete**: Blocked: No documented way to delete an organization (API, CLI, MCP or docs). We deleted every inbox, webhook and key headless; the last key's revoke made the next call 401, but the empty organization stays. ## Does AgentMail have an MCP server, and can an agent connect to it? Yes. AgentMail has an official MCP server. A new agent can get in on its own: in the drive of 14 Oct 2026, our agent signed up, verified and got its key without a person. Through the MCP server, 13 of our agent's 15 attempts worked. ## How does an AI agent get an AgentMail API key? In the drive of 14 Oct 2026, our agent got its AgentMail API key over the API, without a browser or a person. The API key comes in the sign-up response; a scoped key via POST /v0/api-keys worked once the verify had propagated (first try 403 with a clear 'wait up to 5 minutes' fix), was used and revoked. ## Paying the agent way Reaches the checkout: an agent payment on its API, none made yet. The checkout: Plan and price: Developer, $20 a month; A visible CAPTCHA at checkout; Offers x402 payment. Ways to pay it offers an agent: Stripe Projects (on its own domain); x402 (on its own domain · Base, Solana, Avalanche, X Layer, Polygon); MPP (on its own domain · listed or named only); Agentic Commerce Protocol (on its own domain · listed or named only). No card needed to start. What would let agents pay on their own: - Sell the plan or credits over the rail, not only side endpoints: Let the paid call buy what the product sells, so the rail covers the whole product. - Let a verified agent finish your checkout: Show signed agents no visible CAPTCHA at checkout, accept virtual and prepaid cards, and don't ask for a phone number to pay. Ways to pay checked 14 Oct 2026. ## What an agent can do - Sign up, verify and get an API key with two API calls and one emailed code - Create, rename and delete inboxes - Send mail with attachments and read it back - Receive mail and read it by API, WebSocket, polling webhook or IMAP - Reply, reply-all and forward inside a thread - List, label and delete threads and messages - Create, edit, send and schedule drafts - Search messages and threads - Block and allow addresses per inbox - Create scoped API keys and public-key credentials - Read usage, event and rate metrics - Do all core work through the MCP server or the CLI too ## What it cannot do - Delete the account: No documented organization delete; only data and keys can be removed - Buy a plan or reach checkout: Plans are bought in the console, behind a Cloudflare interstitial at sign-in; no billing API - Sign up in the console or use MCP OAuth: Cloudflare Turnstile on sign-up and an interstitial on sign-in - Email more than 3 new recipients in the first hour: Free-plan sending ramp (5 the first day, 10 the first week), a clear 429 with Retry-After - Add a custom domain: Paid plan (Developer) ## Tips for the owner 1. Add an organization delete endpoint 2. Let verified agents through the console check 3. Sell plans through a billing API 4. Give DCR clients the device grant 5. Answer bad keys with a 401 body 6. Serve the MCP server card at /server-card Is this your product? Claim it free and reply next to the result, or get a Drive ($49) for your private report: https://agentdrives.com/pricing?via=md#drive. Fixed something? Order a Drive ($49) any time, or the Monthly drive ($29/month). ## Also in Email infrastructure - [Mailgun](https://agentdrives.com/tools/mailgun.md?via=md): L1 · Needs a person · [AgentMail vs Mailgun](https://agentdrives.com/compare/agentmail-vs-mailgun?via=md) - [Mailtrap](https://agentdrives.com/tools/mailtrap.md?via=md): L1 · Needs a person · [AgentMail vs Mailtrap](https://agentdrives.com/compare/agentmail-vs-mailtrap?via=md) - [Postmark](https://agentdrives.com/tools/postmarkapp.md?via=md): L1 · Needs a person · [AgentMail vs Postmark](https://agentdrives.com/compare/agentmail-vs-postmarkapp?via=md) - [Resend](https://agentdrives.com/tools/resend.md?via=md): L1 · Needs a person · [AgentMail vs Resend](https://agentdrives.com/compare/agentmail-vs-resend?via=md) - [SendGrid](https://agentdrives.com/tools/sendgrid.md?via=md): L1 · Needs a person · [AgentMail vs SendGrid](https://agentdrives.com/compare/agentmail-vs-sendgrid?via=md) ## Tips for agent builders - Follow llms.txt: POST /v0/agent/sign-up with a human_email you can read, then POST /v0/agent/verify with the emailed code. - Wait ~5 minutes after verify before creating scoped keys; the 403 says so. - On the free plan you can mail only 3 new recipients in the first hour; replies to people who wrote to you don't count. - Use Idempotency-Key unquoted (A-Z a-z 0-9 -. _ ~) on sends, client_id on creates. - MCP: send x-api-key to https://mcp.agentmail.to/mcp; the OAuth path hits a Cloudflare check. - Polling webhooks need no public URL but events arrive with a delay; WebSockets are instant. - Errors carry a fix field that names the next step: read it.