# MyAgentMail

> Email + LinkedIn intent infrastructure for AI agents. Real two-way
> email inboxes (SMTP/IMAP, custom domains, WebSocket events,
> workspaces) plus a LinkedIn intent layer (keyword signals,
> engagement signals, watchlist signals, historical search) with
> LLM-driven classification and HMAC-signed webhook delivery.

This file is the canonical entry point for AI agents integrating with
MyAgentMail. Read [SKILL.md](https://myagentmail.com/skills/myagentmail/SKILL.md)
first; it's the single source of truth for capabilities, auth, and
recipes. The OpenAPI spec is the authoritative endpoint reference.

For one-fetch full context, use
[/llms-full.txt](https://myagentmail.com/llms-full.txt) — it inlines
SKILL.md plus every reference doc into a single file.

Base URL: `https://myagentmail.com/v1`. Auth: `Authorization: Bearer tk_...`
(get yours at https://myagentmail.com/dashboard/api-keys).

## Quick start

- [CLI install](https://www.npmjs.com/package/myagentmail-cli): `npm install -g myagentmail-cli && myagentmail login && myagentmail mcp install` — three commands, auto-detects every MCP host on the machine and wires it up. Token-economical for long agent sessions (Claude Code, CI, terminal-based agents).
- [SDK install](https://www.npmjs.com/package/myagentmail): `npm install myagentmail`
- [MCP server install](https://www.npmjs.com/package/myagentmail-mcp): `npx myagentmail-mcp`
  (works with Claude Desktop, Cursor, Windsurf, Cline, any MCP-compatible client)
- [Hello world (curl)](https://myagentmail.com/skills/myagentmail/SKILL.md#quick-start-for-agents): provision an inbox, send mail, listen for replies in 6 lines.

## Authoritative specs

- [SKILL.md](https://myagentmail.com/skills/myagentmail/SKILL.md): primary skill manifest. Auth, key types, capabilities, recipes. Read this first.
- [OpenAPI](https://myagentmail.com/openapi.yaml): every endpoint, every request/response shape. Authoritative.
- [LinkedIn skill reference](https://myagentmail.com/skills/myagentmail/references/linkedin.md): sessions, intent signals (keyword + engagement + watchlist), historical search, multi-session routing, end-to-end recipe.
- [Webhooks reference](https://myagentmail.com/skills/myagentmail/references/webhooks.md): event types, HMAC-SHA256 verification, retry schedule, payload schemas for every event.
- [WebSockets reference](https://myagentmail.com/skills/myagentmail/references/websockets.md): real-time inbox events, subscription model, reconnect semantics.

## Code packages

- [myagentmail](https://www.npmjs.com/package/myagentmail) (npm): typed TypeScript SDK with full coverage of inboxes, messages, drafts, LinkedIn signals, sessions, searches.
- [myagentmail-mcp](https://www.npmjs.com/package/myagentmail-mcp) (npm): Model Context Protocol server. ~25 tools spanning inbox + LinkedIn + signals. Drop into any MCP-compatible client.
- [myagentmail-cli](https://www.npmjs.com/package/myagentmail-cli) (npm): command-line interface. `myagentmail login`, `myagentmail mcp install` (auto-wires Claude Desktop / Cursor / Windsurf / Claude Code), `myagentmail listen --inbox <id> --forward <localUrl>` for webhook tunneling, `myagentmail doctor` for a health-check matrix. Cheaper in agent-token cost than MCP for long sessions because the surface is discovered lazily through `--help`.

## Reference apps

- [myagentmail-outreach-starter](https://github.com/kamskans/myagentmail-outreach-starter): open-source ~3,500 line Next.js app. AI-inferred ICP from a website URL → intent signals → unified leads queue → AI-drafted LinkedIn note + cold email → one-click send. Best example of an end-to-end agent built on MyAgentMail.

## Tutorials

- [Build an Intent-Based LinkedIn Outreach System in an Afternoon](https://myagentmail.com/blog/intent-based-outreach-tutorial): walks through the starter, end-to-end. Discovery → setup → leads → outreach. Covers all three signal kinds.
- [Building a Cold Outreach Agent](https://myagentmail.com/blog/building-cold-outreach-agent): focused tutorial on the email + agent loop without LinkedIn signals.

## Optional

These are useful background but skippable when context-budget-constrained.

- [Why AI agents need email](https://myagentmail.com/blog/why-ai-agents-need-email): conceptual framing of why agents need a real email identity.
- [MyAgentMail vs Cloudflare Email for Agents](https://myagentmail.com/blog/myagentmail-vs-cloudflare-email-for-agents): comparison.
- [WebSocket vs Webhooks for agent email](https://myagentmail.com/blog/websocket-vs-webhooks-agent-email): when to use which.
- [Email deliverability for agents](https://myagentmail.com/blog/email-deliverability-for-agents): SPF/DKIM/DMARC and warmup.

## Conventions this site follows

- Every documentation page is available as markdown (`.md`) at the same URL — append `.md` to any blog or skill URL to get the markdown.
- The OpenAPI spec at `/openapi.yaml` is regenerated on every API release; pin to a tag if you need stability.
- Webhook payloads are HMAC-SHA256 signed using the per-signal `whsec_...` secret returned at signal creation. Verify with `crypto.createHmac("sha256", secret).update(rawBody).digest("hex")` and compare to the `X-MyAgentMail-Signature: v1=<hex>` header.
- All endpoints are versioned at `/v1/...`. Breaking changes ship as `/v2/...` — no silent breakage on `/v1`.