Give every AI agent an email address of its own
An agent inbox is a mailbox, an address on your domain and an API key that can touch only that mailbox. Your agent sends, reads and waits for email over REST, the fm CLI or MCP, and you can watch everything it does in the web inbox.
Works with a domain you already own. One price for your whole team.
- s
signup-bot@orbitlabs.dev
Signs up for services and confirms the verification email
- r
research-agent@orbitlabs.dev
Receives the reports and alerts it subscribes to
- i
invoices-agent@orbitlabs.dev
Collects vendor invoices and pulls out the totals
- o
ops-agent@orbitlabs.dev
Reads alert and status mail from your own services
- t
test@orbitlabs.dev
Throwaway address for agent experiments
Unlimited addresses, catch-all and plus-addressing included.
Why FutureMail works for AI agents
A key that can't wander
Agent keys are pinned to their own mailbox, so a buggy or prompt-injected agent can't read the rest of your company's mail. For tighter limits, give it only the send or read scope, or issue a separate mailbox-pinned key with an expiry.
Wait for the code, not a polling loop
GET /v1/emails/wait long-polls for up to 5 minutes and returns the matching email with extracted.otp and extracted.links, so an agent signing up for a service gets its code in one call.
Remote MCP for Claude and Cursor
Add the Streamable HTTP server at https://api.futuremail.dev/mcp with a Bearer key. Tools such as send_email, search_emails, read_thread and wait_for_email show up in Claude, Cursor or any MCP client.
Scriptable from the shell
The fm CLI (npm i -g futuremail) takes --json on every command, and CODE=$(fm wait --from github.com --otp) drops a one-time code straight into a script.
Start a run the moment mail arrives
Subscribe to email.received with HMAC-signed webhooks, or hold open the Server-Sent Events stream at /v1/events, and kick off your agent as soon as a reply lands.
Humans can read along
Agent mailboxes appear in the same keyboard-first web inbox as your own, so you can check what an agent sent and received, search its threads and step in.
Agent-ready
A signup agent, start to finish
Create an inbox with POST /v1/agents, fm agents create or the create_agent_inbox MCP tool, and store the key it returns, which is shown once. The agent fills in a signup form with its new address, calls wait_for_email filtered by sender, and enters extracted.otp to finish.
- Sign up for a SaaS trial and confirm the emailed code
- Receive vendor invoices and extract the totals for your books
- Reply in-thread to a person who emailed the agent's address
# give an agent its own inbox and a key scoped to it $ fm agents create "Signup bot" \ --address signup-bot@orbitlabs.dev ✓ Created agent Signup bot address signup-bot@orbitlabs.dev scopes send, read # connect Claude, Cursor or any MCP client url https://api.futuremail.dev/mcp header Authorization: Bearer fm_live_…
Set up in an afternoon
- 1
Subscribe
Create an account and start Pro for $20/month. Cancel anytime.
- 2
Connect your domain
Copy the DNS records we generate into your registrar. FutureMail checks them live and tells you when you're set.
- 3
Create your addresses
Add the addresses you need, point them at shared or individual mailboxes, and invite your team.
- 50,000 recipients a month
- Unlimited custom domains
- Unlimited addresses & aliases
- Agent inboxes with scoped API keys
- REST API, CLI & remote MCP server
- Webhooks, templates & scheduled send
- Open & click tracking
- Team roles & audit log
Paid only, cancel anytime. New accounts can send to up to 1,000 recipients a day for their first 14 days; receiving is never limited. Pricing details
Common questions
How do I give an AI agent its own email address?
Create an agent inbox in Settings → Agents, with POST /v1/agents, with fm agents create, or with the MCP tool create_agent_inbox. You get a mailbox, an address on your domain and an API key that only works on that mailbox.
Can an agent read my other mailboxes?
Not with its agent key. Agent keys are always pinned to the agent's own mailbox, so it can read and send only from its own inbox, and it can never manage members or billing.
Does FutureMail include the AI model?
No. FutureMail provides the inbox, address, API, CLI and MCP server. The agent is yours: Claude, Cursor or your own code calling the API.
Is there a local MCP server I can run myself?
No. FutureMail runs one remote MCP server over Streamable HTTP at https://api.futuremail.dev/mcp. Clients that support remote HTTP servers, such as Claude Code and Cursor, connect with an Authorization: Bearer header.
How much does FutureMail cost? Is there a free plan?
FutureMail is paid-only: Pro is $20/month and includes 50,000 recipients a month, unlimited domains and addresses, agent inboxes, the API and MCP server. There is no free tier or trial. Limits are hard caps, so there are never overage charges and no per-seat pricing. Charged in USD; Stripe may show your local currency and add applicable tax. Need more volume? Email support@futuremail.dev.
More answers in the general FAQ.
Related use cases
AI assistants
Connect Claude or any MCP client to your FutureMail inbox to triage, search, summarize and draft replies, or give your assistant its own address people can cc.
QA and testing
Test signup, password reset and magic-link emails end to end: a unique plus-address per run, wait_for_email or fm wait --otp in CI, and simulated bounces.
SaaS products
Send receipts, password resets and sign-in codes from your SaaS via REST or a Resend-compatible API, and get a real inbox so customer replies reach your team.