Send real email from a Python AI agent in 20 lines

Most AI agents today live behind a chat UI. That's fine for demos, but the moment you want your agent to participate in real work — coordinating with customers, scheduling meetings, responding to vendor outreach — you need the inbox everyone else uses. Email.

This post walks through the shortest path to a Python agent that actually sends and receives email, with verified sender identity, conversation threading, and real-time delivery. No SMTP server, no public webhook URL (if you don't want one), no DNS.

Get an email address for your agent

Sign up at e2a.dev, pick a slug like research-assistant, and you'll get research-assistant@agents.e2a.dev. Auto-verified, ready to receive mail immediately. If you prefer your own domain, you can register it instead (MX + TXT record), but the shared domain is faster for experiments.

Install the Python SDK:

pip install e2a

Send an email

Three lines:

from e2a.v1 import E2AClient

client = E2AClient(api_key="e2a_...")

client.send(
    to=["alice@example.com"],
    subject="Quick question",
    body="Hi Alice — I'm your scheduling assistant. Got a second for a quick sync?",
)

That's it. The message goes out signed with your domain's DKIM, shows up in Alice's inbox looking like a normal email, and her reply will route back to your agent via e2a.

Receive replies

Two modes, depending on whether your agent has a public URL:

Cloud mode (webhook) — when Alice replies, e2a POSTs the email to your server:

@app.post("/webhook")
async def inbox(request):
    email = client.parse(await request.body())
    print(f"{email.sender}: {email.subject}")
    await email.reply(f"Got it. Does Thursday 2pm work?")
    return {"ok": True}

Local mode (WebSocket) — no public URL needed. Great for running agents on your laptop or behind a firewall:

from e2a.v1 import AsyncE2AClient

async with AsyncE2AClient(api_key="e2a_...") as client:
    async for email in client.listen("research-assistant@agents.e2a.dev"):
        print(f"{email.sender}: {email.subject}")
        await email.reply("Got it, on it.")

listen() holds a WebSocket, receives a lightweight notification when mail arrives, fetches the full message via REST, and yields an AsyncInboundEmail object you can reply to inline.

Thread conversations across turns

Multi-turn threading is the difference between an agent that sends isolated emails and one that maintains context. Pass conversation_id on each outbound and you'll see it back on subsequent inbounds in the same thread:

async for email in client.listen("research-assistant@agents.e2a.dev"):
    convo_id = email.conversation_id or new_id()

    draft = await compose_reply(email, history=load_history(convo_id))
    await email.reply(draft, conversation_id=convo_id)

Works across humans replying from Gmail and other e2a agents replying via the platform. First contact from a human arrives with conversation_id=None — assign one yourself, and every reply in that thread will carry it.

Verify the sender

Every inbound includes signed auth headers showing SPF/DKIM results. Use them to decide how much to trust the message:

if email.is_verified:
    # SPF/DKIM passed; sender domain is authenticated
    handle(email)
else:
    # Could be a spoof — route to quarantine or ask for confirmation
    flag_for_review(email)

The signature is HMAC-SHA256 over a canonical string; you can verify it yourself against your webhook secret if you want defense in depth.

What you can build from here

The three primitives — send, listen/webhook, and conversation_id threading — cover most agent-in-your-inbox use cases. A few ideas:

• A scheduling assistant that you CC on meeting threads; it coordinates with the other side and books calendar holds.
• A customer-support triage agent that handles first replies and escalates to humans when confidence drops.
• A voice agent that sends follow-up emails after calls, then keeps the thread going when the person replies.
• A procurement bot that chases vendors for quotes and aggregates responses.

Anything that can receive email and take action can now have an agent attached to it. The Python SDK README and API reference cover attachments, CC/BCC, reply-all, and the auth header details if you want to go deeper.