MyAgentMail

MyAgentMail gives AI agents two things they almost never get out of the box: their own real email inboxes, and a LinkedIn intent layer.

Most agent frameworks can send email by piggybacking on a customer's Gmail OAuth, but they can't receive — and OAuth tokens expire mid-workflow, get revoked when the human reads a security email, or simply don't exist on day one of a SaaS product. They can scrape LinkedIn, but they can't sit in the background watching for buying signals and fire a webhook the moment one lands. We built MyAgentMail because every serious outbound, research, or recruiting agent we tried to ship hit those two walls.

Real two-way email for agents

Inboxes that behave like real inboxes — not webhooks pretending to be email.

• Send + receive + thread. Every outbound message carries proper Message-ID + In-Reply-To headers, so when a recipient replies, the response lands on the same thread automatically. No fragile subject-line matching. Read whole threads via GET /v1/threads/:id, reply via POST /v1/messages/reply, forward, mark-read, delete — all the verbs you'd expect from a real mail client, exposed as a JSON API.

• Custom domains. Register your own sending domain through the API; we generate the DKIM/SPF/DMARC records and verify them in-platform. Customers see emails from @yourdomain.com, not @some-vendor-subdomain.com.

• SMTP + IMAP credentials per inbox. When you need to integrate with a legacy tool or let a human jump into the same mailbox alongside the agent, every inbox ships with standard mail-server credentials. The agent and the human share state.

• Attachments. Inbound attachments are extracted, scanned, and exposed via GET /v1/attachments/:id. Outbound attachments accepted on send.

• Inbound webhooks. Subscribe to message.received, message.replied, message.bounced, message.complained and react in your own code. Reply-detection is the foundation of any cadence that stops itself when a human responds.

• Drafts. Iterative composition — create, update, send. Useful when the agent wants a human to approve before sending, or when message generation runs in multiple LLM passes.

• Deliverability that's actually managed. We run our own SES infrastructure on a dedicated send.myagentmail.com subdomain (apex stays clean for receiving), enforce sender warmup, and surface complaint + bounce rates so you can tell before a customer's reputation is in trouble.

LinkedIn intent layer

The other half of outbound that nobody else solves at API level.

• Continuous monitoring. Write a firing rule in plain English ("flag founders complaining about cold email", "anyone hiring a head of growth in NYC fintech"). We poll continuously, match posts against the rule with an LLM, and fire a webhook the instant a match lands.

• Quota that grows with you. Distribution across every LinkedIn account you connect — daily monitoring quota scales linearly with each session, not with how much you pay us.

• Historical search. Same rule language, but run once across the past 24 hours, week, or month. Stateless, synchronous — for backfills, account research, or one-shot intent sweeps.

• Inline 2FA wait. When an agent signs into a LinkedIn account that triggers a security code, the API holds open a WebSocket and resolves the moment your operator submits the code. No polling loops, no brittle retry hacks.

• Connect, DM, post, comment, search profiles, list connections. Full programmatic access to the same actions a human performs in the browser, with per-account daily action caps to keep the session safe.

Cadences — the glue

Multi-step outreach sequences that combine both channels.

linkedin_connect → wait 2d → linkedin_message → wait 3d → email_fallback → wait 5d → email_followup

Each step's message body can come from your own webhook, so wording stays in your model and your voice. Reply detection across both channels stops the cadence the moment a human responds — no double-sending after a reply.

Multi-tenant via workspaces

One MAM tenant can spawn isolated workspaces. LinkedIn session pools, intent signal quotas, inboxes, and cadences all scope to a workspace. The model SaaS products built on top of MAM use (Outrly, Delegated, and others): one MAM customer at the platform layer, N customer workspaces underneath, with workspace-scoped wk_ keys that can't see across the boundary.

How you use it

npm install myagentmail@latest for the TypeScript SDK, or hit the raw HTTP API (X-API-Key header, public OpenAPI spec). Every primitive — inboxes, threads, signals, sessions, cadences, workspaces — has an MCP server too, so Claude / Cursor / any agent runtime can call them directly without you writing wrappers.

What it isn't

Not a marketing automation platform. Not a CRM. Not an end-user email client. It's the primitives — inboxes, intent signals, cadences, sessions — that other products are built on top of.

Pricing

Two stacks, mix and match.

Email — pay for the volume you send.

• Starter $5/mo: 2 inboxes, 6K emails/mo, 2 custom domains

• Developer $15/mo: 10 inboxes, 20K emails/mo, 10 custom domains, workspace-scoped keys

• Startup $199/mo: 200 inboxes, 150K emails/mo, 200 custom domains, reseller workspaces

LinkedIn intent layer — pay for the accounts you monitor.

• Solo $29/mo: 2 connected accounts, 100 actions/day each

• Team $99/mo: 15 connected accounts, 500 actions/day each

Cadences and webhooks are included in every tier. One Stripe customer covers both stacks.