API-first email for AI agents

Your agents need
their own inboxes

Create email inboxes with one API call. Your agents send, receive, reply, and forward — with threading, webhooks, real-time WebSocket events, semantic search, ephemeral TTL inboxes, long-poll wait-for-message, sub-addressing via plus-tags, and a native MCP server for Claude Code and Codex. Hosted in Singapore.

Get your API key Free tier · No credit card · PDPA compliant

AgentMail Console

agentmail.cyberforge.one/dashboard

Inboxes

Email addresses for your AI agents

+ New Inbox

support-agent@acme.com(Customer Support AI)

Last message 2 min ago

Copy

intake@acme.com(Document Intake Bot)

Last message 14 min ago

Copy

scheduler@acme.com(Meeting Scheduler)

Last message 1 hr ago

Copy

3 active inboxes1,247 emails this month99.8% delivery rate

How it works

Create inboxes via API

One POST request creates a fully functional email inbox. DKIM, SPF, and DMARC configured automatically. No DNS wait, no manual setup.

inbox = client.inboxes.create(
    username="support-agent",
    domain="yourdomain.com"
)
# → support-agent@yourdomain.com is live

Send, reply, and forward

Your agents send, reply, and forward emails programmatically. One-call reply handles Re: prefixes and In-Reply-To threading automatically. Use plus-tag sub-addressing to route workflows through a single inbox — send from support+billing@ and filter by sub-address. Receive new mail via real-time webhooks — HMAC-signed, per-inbox scoped — or connect via WebSocket for instant push events without a public endpoint.

Search, extract, and process

Search messages by keyword, sender, direction, or time window — agents find OTPs or confirmations in milliseconds. Or use the wait-for-message endpoint to block until a reply arrives — no polling loop needed. Filter by date range (after/before), filter for messages with attachments, and download attachment files directly via API. Or integrate natively via MCP — no HTTP code needed.

Why AgentMail

Built from Singapore,
for Asia-Pacific

Most email APIs route through US data centres. That means 200ms+ latency, US-only data residency, and support during California business hours. We built the alternative.

<50ms

API response time from Singapore

Your APAC agents respond in real-time. No round-trip to Virginia.

PDPA compliant by default

Data stays in APAC. Meet Singapore PDPA, Thailand PDPA, and MAS TRM requirements without extra work.

SGD invoicing

No FX fees. Local currency for APAC businesses.

APAC timezone engineering

Issues resolved during your business hours, not California's.

Use cases

What teams build with AgentMail

Customer service agents

Each agent gets its own inbox. Customers email support@, your AI handles it with full conversation threading.

Document intake

Receive invoices, contracts, and receipts via email. Agents extract structured data and route it.

Agent authentication

Give each agent a verified email for signing up to services, receiving OTPs, and identity verification.

Mutable Inboxes

Rename and relabel inboxes on the fly. Update displayName without recreating — partial updates, null to clear.

Usage Stats API

One call returns your full account picture: inbox count, message volumes, unread/quarantine counts, and top-5 busiest inboxes.

Inbox Purge

Reset a disposable inbox in one DELETE call — bulk-delete all messages without recreating the inbox. Optional ?before=ISO8601 for selective purging by date.

Global Cross-Inbox Search

Search across all your inboxes simultaneously. Find any message without knowing which inbox it arrived in — essential for multi-agent setups.

Reply & Forward API

One-call reply or forward — no manual header wrangling. POST .../reply and the API handles Re: prefix, In-Reply-To, and thread continuity automatically.

Ephemeral Inboxes (TTL)

Set ttlSeconds at creation and the inbox self-destructs automatically. Ideal for single-task agent workflows that should never leave residual state.

Native MCP Tools

Claude Code, Codex, and Gemini CLI can call send_email, list_threads, and search_messages as native tools. No HTTP boilerplate. 16 tools available out of the box.

Agent Identity

Inboxes can carry agentName and agentModel metadata for multi-agent fleet visibility. Know exactly which agent sent what — essential for audit trails in production deployments.

Sub-Addressing (Plus Tags)

Route workflows through a single inbox using plus-tags. Send from support+billing@acme.com via the fromTag parameter, then filter incoming messages by subAddress. Every message response includes its resolved sub-address — zero extra inboxes needed.

Direct Message Listing

Skip the thread abstraction. GET /inboxes/:id/messages returns paginated messages directly — newest first, with unread filtering. The fastest way to process an inbox without navigating thread IDs.

Wait-for-Message (Long-Poll)

Eliminate polling loops. GET /inboxes/:id/wait blocks until a new email arrives (or timeout). Your agent awaits OTPs, magic links, and replies with a single call — no while-loop, no wasted quota.

Webhook DLQ: Retry & Replay

When your endpoint goes down, failed webhooks land in a Dead Letter Queue with automatic exponential backoff retries — 5 attempts over 25 hours. Inspect payloads, manually replay individual events or replay-all in one call. 72-hour retention with auto-cleanup.

Idempotency Keys

Send an Idempotency-Key header on create/send requests. Retries after timeouts return the cached response instead of creating duplicates. Essential for autonomous agents with unreliable networks.

Per-Agent Usage Analytics

Track emails sent, received, and webhook events per inbox. GET /api/v1/analytics/inboxes returns per-agent breakdowns — know exactly which agents are driving volume before you hit a limit.

Configurable Send Caps

Set daily_send_cap, monthly_send_cap, and alert_threshold_pct per inbox via PATCH /inboxes/:id. Prevent runaway agents from burning through quotas — get alerted at 80%, hard-stopped at 100%.

Namespace-Based Billing & Quotas

Group inboxes into namespaces by agent, customer, or team. Set per-namespace send caps, inbox limits, and receive quotas — enforced at the API level (429 on breach). Pull charge-through billing line items per namespace for direct pass-through to your Stripe or billing system.

Real-Time Events

Connect via WebSocket to GET /api/v1/ws. Receive message.received and message.quarantined events the instant they occur. No polling. No public endpoint required. Ideal for agents running inside private networks or local dev environments.

API Reference

Everything is an API call

No SDK required. Every feature is a REST endpoint. Authenticate with a Bearer token, send JSON, get JSON back.

POST/api/v1/inboxes

Create a new inbox for your agent

Request

{
  "username": "support-agent",
  "displayName": "Customer Support AI"
}

Response

{
  "ok": true,
  "data": {
    "id": "inb_abc123",
    "address": "support-agent@yourdomain.com",
    "dkimSelector": "am1710489600",
    "status": "active"
  }
}

POST/api/v1/inboxes/:id/send

Send an email from your agent's inbox. Use fromTag to send from a plus-tagged sub-address (e.g. support+billing@).

Request

{
  "to": ["customer@example.com"],
  "subject": "Re: Your request",
  "bodyText": "We've processed your order.",
  "fromTag": "billing",
  "replyToMessageId": "msg_xyz789"
}

Response

{
  "ok": true,
  "data": {
    "id": "msg_def456",
    "threadId": "thr_abc123",
    "subAddress": "billing",
    "status": "sent"
  }
}

GET/api/v1/inboxes/:id/threads

List conversation threads with messages

Response

{
  "ok": true,
  "data": {
    "threads": [{
      "id": "thr_abc123",
      "subject": "Order #1234",
      "messageCount": 4,
      "unreadCount": 1,
      "participants": ["agent@acme.com", "customer@example.com"]
    }]
  }
}

GET/api/v1/attachments/:id

Download an email attachment — stream the file directly. Use ?inline=true to render in browser. Agents can process invoices, contracts, and PDFs received via email.

Response

// Binary file stream
// Content-Type: application/pdf
// Content-Disposition: attachment; filename="invoice.pdf"
// Content-Length: 48291
// Use GET /attachments/:id/meta for JSON metadata

GET/api/v1/inboxes/:id/messages/search

Search messages with date-range + attachment filters. Critical for OTP workflows: find emails by time window. New params: after, before (ISO 8601), hasAttachments (true/false).

Response

// GET ?q=OTP&after=2026-03-15T22:00:00Z&hasAttachments=false
{
  "ok": true,
  "data": {
    "messages": [{ "id": "msg_...", "subject": "Your OTP", "receivedAt": "..." }],
    "query": { "after": "2026-03-15T22:00:00Z", "hasAttachments": false }
  }
}

GET/api/v1/inboxes/:id/messages

Direct paginated message listing — skip the thread abstraction. Returns all messages for an inbox, newest first. Supports ?limit, ?cursor, ?unread=true, and ?subAddress=billing filtering.

Response

// GET /api/v1/inboxes/:id/messages?limit=5&subAddress=billing
{
  "ok": true,
  "data": {
    "messages": [{ "id": "msg_...", "subject": "Invoice #1234", "subAddress": "billing", "readAt": null }],
    "nextCursor": "msg_xyz..."
  }
}

GET/api/v1/inboxes/:id/wait

Long-poll wait for new message — blocks until an email arrives or timeout expires. Eliminates polling loops. Pass ?timeout=30 (5–60s) and ?after=<ISO8601> to only trigger on new messages.

Response

// GET /api/v1/inboxes/:id/wait?timeout=30&after=2026-03-23T23:00:00Z
// Blocks up to 30s, returns immediately on new message:
{ "ok": true, "data": { "id": "msg_...", "subject": "OTP: 847291" } }

// Or on timeout:
{ "ok": true, "data": null, "timeout": true }

GET/api/v1/analytics

Account-wide usage analytics — total emails sent/received today and this month, per-inbox event breakdown, and webhook delivery counts. One call gives you the full picture.

Response

{
  "ok": true,
  "data": {
    "period": "today",
    "totals": { "sent": 142, "received": 89, "webhooks_fired": 231 },
    "byInbox": [
      { "inboxId": "inb_abc", "address": "support@acme.com", "sent": 98, "received": 44 }
    ]
  }
}

GET/api/v1/analytics/inboxes/:id

Per-inbox analytics — daily and monthly send/receive counts, current cap utilisation, and alert threshold status. Use to build usage dashboards or drive auto-scaling logic.

Response

{
  "ok": true,
  "data": {
    "inboxId": "inb_abc123",
    "daily": { "sent": 42, "received": 18, "cap": 500, "pct_used": 8.4 },
    "monthly": { "sent": 1204, "received": 531, "cap": 10000, "pct_used": 12.0 },
    "alert_threshold_pct": 80,
    "alert_triggered": false
  }
}

POST/api/v1/webhooks

Request

{
  "url": "https://yourapp.com/webhook",
  "events": ["message.received", "message.sent"],
  "inboxId": "inb_abc123"
}

Response

{
  "ok": true,
  "data": {
    "id": "wh_abc123",
    "secret": "whsec_...",
    "status": "active"
  }
}

GET/api/v1/webhooks/:webhookId/dlq

List dead letter queue events for a webhook — failed and pending deliveries. Paginated. Query: status (failed|pending), limit, cursor.

Response

{
  "ok": true,
  "data": {
    "events": [{
      "id": "wde_abc123",
      "webhookId": "wh_xyz",
      "eventType": "message.received",
      "payload": { ... },
      "status": "failed",
      "attempts": 3,
      "lastError": "Connection refused",
      "nextRetryAt": "2026-04-13T12:00:00Z",
      "createdAt": "2026-04-12T10:00:00Z"
    }],
    "nextCursor": "wde_xyz..."
  }
}

POST/api/v1/webhooks/:webhookId/dlq/:eventId/replay

Manually replay a single failed event. Re-fires the original payload immediately, bypassing backoff schedule.

Response

{
  "ok": true,
  "data": {
    "eventId": "wde_abc123",
    "status": "replayed",
    "deliveryId": "wdl_xyz"
  }
}

POST/api/v1/webhooks/:webhookId/dlq/replay-all

Replay all failed events for a webhook in one call. Each fires immediately with backoff schedule reset.

Response

{
  "ok": true,
  "data": {
    "queued": 4,
    "message": "4 events queued for replay"
  }
}

GET/api/v1/ws

WebSocket real-time event stream — connect once and receive message.received and message.quarantined events the moment they occur. No polling. No public webhook endpoint required. Authenticate via Authorization header or ?token= query param. Perfect for agents in private networks or local environments.

Response

// Connect: wss://agentmail.cyberforge.one/api/v1/ws
// Auth header: Authorization: Bearer am_live_YOUR_KEY
// Or query param: ?token=am_live_YOUR_KEY

// Incoming events (JSON, one per message):
{ "event": "message.received", "data": { "id": "msg_...", "inboxId": "inb_...", "subject": "Your OTP", "from": "noreply@service.com", "receivedAt": "2026-03-31T00:12:34Z" } }
{ "event": "message.quarantined", "data": { "id": "msg_...", "inboxId": "inb_...", "reason": "spam_score_exceeded", "receivedAt": "2026-03-31T00:12:35Z" } }

Also available: GET /inboxes · DELETE /inboxes/:id · PATCH /inboxes/:id · GET /api/v1/account/stats · GET /api/v1/analytics · GET /api/v1/analytics/inboxes · PATCH /messages/:id/read · DELETE /inboxes/:id/messages · GET /messages/search · POST /inboxes/:id/messages/:msgId/reply · POST /inboxes/:id/messages/:msgId/forward · GET /api/v1/ws (WebSocket event stream) · GET /inboxes/:id/quarantine · POST /auth/keys · GET /attachments/:id · GET /attachments/:id/meta · POST /mcp (MCP server — JSON-RPC 2.0) · GET /api/v1/namespaces · POST /api/v1/namespaces · PATCH /api/v1/namespaces/:slug · DELETE /api/v1/namespaces/:slug · GET /api/v1/namespaces/:slug/billing

OpenClaw Integration

Give your AI assistant
a real email address

AgentMail integrates natively with OpenClaw — the open-source AI assistant platform. Install the skill, and your agents can send, receive, and manage email without writing any code.

setup.sh

# 1. Download the AgentMail skill
curl -sL https://agentmail.cyberforge.one/skill \
  -o ~/.agents/skills/agentmail/SKILL.md \
  --create-dirs

# 2. Add your API key
openclaw config set agentmail.apiKey am_live_YOUR_KEY

# 3. Restart to load the skill
openclaw gateway restart

# Done. Your agents can now use email:
# "Create an inbox for my support agent"
# "Send an email to customer@example.com"
# "Check new emails in the intake inbox"
# "Reply to the latest support thread"

One skill file teaches your agent the full API. Add your key, restart, done.

Automated customer support

Your OpenClaw agent monitors an inbox, reads new emails, drafts responses using your LLM, and replies — all without human intervention. Escalates complex issues to your team.

Document processing pipeline

Emails with invoices, contracts, or receipts arrive at an inbox. OpenClaw extracts data, routes to the right workflow, and sends a confirmation reply.

Multi-agent coordination

Each agent in your OpenClaw setup gets its own email identity. They email external services, receive OTPs, and authenticate — just like a human teammate would.

Scheduled reports & alerts

Use OpenClaw cron jobs to check inboxes on a schedule. Process incoming data, compile summaries, and email reports to stakeholders automatically.

MCP Integration

Native MCP server for AI coding agents

Claude Code, Codex, Gemini CLI, and any MCP client can use AgentMail as a tool without writing HTTP code. One config entry, zero boilerplate. AgentMail is the only email API with Tools + Resources + Prompts — the most MCP-native email API for AI agents.

.mcp.json

{
  "mcpServers": {
    "agentmail": {
      "url": "https://agentmail.cyberforge.one/mcp",
      "headers": {
        "Authorization": "Bearer am_live_YOUR_KEY"
      }
    }
  }
}

Add this to your .mcp.json and your coding agent gets 16 email tools immediately — no HTTP client, no boilerplate.

MCP Resources3 resources

Inbox messages as typed ResourcesThread listings as typed ResourcesAccount metrics as typed Resources

MCP Prompts3 prompts

process-inbound-supportfollow-up-quotesummarize-thread

MCP Tools18 tools

list_inboxescreate_inboxget_inboxupdate_inboxdelete_inboxsend_emailget_messagelist_messagesreply_to_messageforward_messagesearch_messagessearch_inboxget_threadlist_threadsget_account_statspurge_inboxdownload_attachmentlist_webhooks

Only AgentMail has Tools + Resources + Prompts

The MCP server at POST /mcp speaks JSON-RPC 2.0. Authentication is the same Bearer token you use for the REST API. 16 tools, 3 Resources (inbox messages, threads, account metrics), and 3 pre-built workflow Prompts (process-inbound-support, follow-up-quote, summarize-thread) — making AgentMail the most MCP-native email API for AI agents.

Pricing

Start free, pay when you scale

No credit card required. No hidden fees. Upgrade when your agents outgrow the free tier.

Free

For prototyping

$0/mo

3 inboxes
3,000 emails/month
Webhook management API
Webhook delivery logs
Idempotency keys
Message search + date filters
Attachment download API
Community support

Start free

Pro

For production

$29/mo

50 inboxes
50,000 emails/month
Custom domains + DKIM
Message search + semantic search + date filters
Attachment download API
Webhook management + delivery logs
Idempotency keys
Priority support

Get started

Enterprise

Custom volume

Custom

Unlimited inboxes
Dedicated IPs
APAC data residency SLA
SSO / SAML
Dedicated support

Talk to us

FAQ

Common questions

Ready to give your agents email?

One API call. One inbox. Start in under two minutes.

Get your API key

Your agents needtheir own inboxes

Inboxes

Create inboxes via API

Send, reply, and forward

Search, extract, and process

Built from Singapore,for Asia-Pacific

What teams build with AgentMail

Customer service agents

Document intake

Agent authentication

Mutable Inboxes

Usage Stats API

Inbox Purge

Global Cross-Inbox Search

Reply & Forward API

Ephemeral Inboxes (TTL)

Native MCP Tools

Agent Identity

Sub-Addressing (Plus Tags)

Direct Message Listing

Wait-for-Message (Long-Poll)

Webhook DLQ: Retry & Replay

Idempotency Keys

Per-Agent Usage Analytics

Configurable Send Caps

Namespace-Based Billing & Quotas

Real-Time Events

Everything is an API call

Give your AI assistanta real email address

Automated customer support

Document processing pipeline

Multi-agent coordination

Scheduled reports & alerts

Native MCP server for AI coding agents

Start free, pay when you scale

Free

Pro

Enterprise

Common questions

Ready to give your agents email?

Your agents need
their own inboxes

Built from Singapore,
for Asia-Pacific

Give your AI assistant
a real email address