# Sendmux Full Context > Extended LLM context for the public Sendmux website. Generated from `/llms.txt` plus the maintained full-context source. Source llms.txt SHA-256: 89ea6d7ed5bc8890ce4a501cd66a5435074fd97da823ec08fafe8749c60ddc78 Source content SHA-256: ede9d23b5068118abb4a371771a1934ad5c2168337288854188eacb6774a8b69 ## Curated index from /llms.txt # Sendmux > Sendmux is the email API for AI agents and SaaS platforms. It combines multi-provider outbound sending, agent inboxes, scoped APIs, webhooks, live mailbox events, and usage-based billing. Sendmux is built for teams that need agents to send, receive, route, and process email without stitching together a sending provider, a mailbox provider, parsing code, webhook delivery, separate credentials, and separate billing. Current public surfaces include the marketing site, Mintlify documentation, Sending API OpenAPI spec, Management API OpenAPI spec, SDKs, CLI, Doc Search MCP, MCP, and A2A. ## Website - [Homepage](https://sendmux.ai/): Product overview for the email API for AI agents. - [Product overview](https://sendmux.ai/product/): Hub for sending, inbound mailboxes, and platform controls. - [Sending API and SMTP relay](https://sendmux.ai/product/sending/): Multi-provider outbound routing, delivery groups, quotas, health checks, failover, HTTP sending, SMTP submission, and idempotency. - [Agent inbox API](https://sendmux.ai/product/inboxes/): Dedicated mailboxes for agents with REST, IMAP, SMTP submission, clean content endpoints, attachments, sender rules, webhooks, and Server-Sent Events. - [Platform API controls](https://sendmux.ai/product/platform/): Team isolation, scoped API keys, signed webhooks, Server-Sent Events, logs, usage controls, and code examples. - [Pricing](https://sendmux.ai/pricing/): Free includes two mailboxes and 50 accepted recipients a day across all outbound routes. Pro is $7 per team each month, plus per-event usage. - [Free email tools](https://sendmux.ai/tools/): Browser-based tools for debugging email after the fact, starting with a header analyzer. No account needed. - [Email header analyzer](https://sendmux.ai/tools/email-header-analyzer/): Paste a raw header block to trace the delivery path in delivery order, the delay at every hop, ARC forwarding chains, and the SPF, DKIM and DMARC results the receiving server recorded. Parses entirely in the browser; nothing is uploaded. - [Blog](https://sendmux.ai/blog/): Technical guides for sending, receiving, authentication, deliverability, and agent email infrastructure. - [Contact](https://sendmux.ai/contact/): Sales and support contact paths. - [Email API for AI agent builders](https://sendmux.ai/solutions/ai-agent-builders/): Audience landing for AI agent builders: real agent mailboxes, authorised sending via MCP and the Sending API, clean reply content, and live mailbox events. - [Email API for multi-tenant SaaS platforms](https://sendmux.ai/solutions/platform-builders/): Audience landing for SaaS and platform builders: one isolated Pro team per customer, mailbox-scoped API keys, signed webhooks, and no per-mailbox fees. - [Enterprise email API for AI agent fleets](https://sendmux.ai/solutions/enterprise/): Sales-led landing for organisations running fleets of AI agents: isolated teams, role-based access, scoped credentials, SHA-256 hashed keys and HMAC-SHA256 signed webhooks, and load-tested scale. Dedicated infrastructure, custom SLAs, priority support, and a security review come by enterprise agreement; SSO/SAML and compliance certifications are on the roadmap. - [Email for AI agents, provisioned by agents](https://sendmux.ai/solutions/for-ai-agents/): Agent-facing landing for agents with no credential to start. `sendmux agent:register ` creates an @myagent.mx inbox with a durable read credential that does not expire and a 500 MiB storage limit. `sendmux agent:invite-owner owner@example.com --profile ` invites the owner. Owner acceptance and sending approval guarantee at least 5 GiB of inbox storage before sending uses a one-hour delegated token. Revoking sending does not itself change the current inbox storage allocation. - [Email for AI customer support agents](https://sendmux.ai/use-cases/customer-support-agents/): Use-case page for AI customer support agents: a dedicated mailbox that receives customer email, reads it as clean JSON with quoted history and signatures stripped, replies from the same address, and applies keyword labels over REST or MCP, with sender rules and usage-based pricing. - [Email API for AI sales outreach agents](https://sendmux.ai/use-cases/sales-outreach-agents/): Use-case page for AI sales outreach agents: route outbound through your own providers with per-provider quotas, weighted distribution, and failover so sending reputation stays on your own identities, then catch every reply in the agent's mailbox, thread it, read it clean, and label intent over REST or MCP. Sendmux is the send-and-reply layer; no warmup, sequencing, or campaign builder. - [Email for AI scheduling agents](https://sendmux.ai/use-cases/scheduling-agents/): Use-case page for AI scheduling agents: a dedicated mailbox that catches the booking request, reads the proposed times as clean JSON with the booking link extracted, threads the whole negotiation, confirms with an agent-built .ics, coordinates several attendees, and reminds or reschedules over REST or MCP. Sendmux is the email layer; the agent and its own calendar do the scheduling. - [Email for AI monitoring agents](https://sendmux.ai/use-cases/monitoring-agents/): Use-case page for AI monitoring agents: a dedicated mailbox that catches every inbound alert over a signed webhook or a live event stream, de-dupes repeat events for 24 hours, filters trusted senders with allowlist rules, reads each alert as clean content, and escalates with an idempotent send so a retried page fires once. Sendmux is the email rail; the agent's own model classifies and decides who to page. ## Developer reference - [Sendmux documentation](https://sendmux.ai/docs): Guides and API references. - [Quickstart](https://sendmux.ai/docs/guides/quickstart): Create a team, add a mailbox, send a test email, and check logs. - [Sending API reference](https://sendmux.ai/docs/sending-api/introduction): HTTP sending API reference. - [Management API reference](https://sendmux.ai/docs/api/introduction): Team, provider, domain, mailbox, webhook, billing, log, and metric API reference. - [Mailbox API reference](https://sendmux.ai/docs/mailbox-api/introduction): Mailbox-scoped API reference for agent mailboxes. - [Send by HTTP](https://sendmux.ai/docs/guides/sending-via-http): Queue single or batch email requests. - [Send via SMTP](https://sendmux.ai/docs/guides/sending-via-smtp): SMTP submission guide for existing clients and agent frameworks. - [Webhooks](https://sendmux.ai/docs/guides/webhooks-setup): Subscribe to delivery and mailbox events. - [Verify webhook signatures](https://sendmux.ai/docs/guides/webhooks-verify-signatures): Verify HMAC-SHA256 webhook deliveries. - [Idempotency](https://sendmux.ai/docs/guides/idempotency): Retry mutating requests without duplicate sends. - [SDKs](https://sendmux.ai/docs/sdks): Package-managed TypeScript, Python, Go, PHP, Ruby, and Rust SDKs for Sending, Mailbox, and Management APIs. - [CLI](https://sendmux.ai/docs/cli): Install `@sendmux/cli` to register agent inboxes, invite owners, and run Management, Mailbox, and Sending workflows. - [MCP servers](https://sendmux.ai/docs/guides/mcp): Connect AI tools to the documentation MCP for docs search or the product MCP for authorised Sendmux tools. - [Agent access service document](https://sendmux.ai/auth.md): Machine-readable entry point for CLI-first agent registration, durable inbox reads, owner invitations, one-hour delegated sending after approval, HTTP fallback, and full revocation. - [Agent Skills guide](https://sendmux.ai/docs/guides/agent-skills): Install the full Sendmux skill pack with `npx skills add Sendmux/skills`, or install focused Sendmux skills. - [Agent skills](https://sendmux.ai/.well-known/agent-skills/sendmux-getting-started/SKILL.md): Official Sendmux agent skills for setup, auth, agent access, sending, mailbox, management, MCP, and CLI. Install the pack with `npx skills add Sendmux/skills`; the full set is indexed at https://sendmux.ai/.well-known/agent-skills/index.json. - [A2A Agent Card](https://sendmux.ai/.well-known/agent-card.json): A2A 1.0 discovery metadata for deterministic, OAuth-authorised Sendmux mailbox, management, and sending operations. - [API catalog](https://sendmux.ai/.well-known/api-catalog): Linkset JSON catalog of Sendmux API endpoints, docs, and OpenAPI specs. - [Web Bot Auth key directory](https://sendmux.ai/.well-known/http-message-signatures-directory): JWKS for verifying signed Sendmux bot and agent requests. - [Sending API OpenAPI 3.1 spec](https://smtp.sendmux.ai/api/v1/openapi.json): Machine-readable spec for sending endpoints. - [Management API OpenAPI 3.1 spec](https://app.sendmux.ai/api/v1/openapi.json): Machine-readable spec for management and mailbox endpoints. ## Pricing - Free: $0 per month, two mailboxes, $1 usage credit, fixed limits, and 50 provider-accepted recipients per UTC day across all outbound routes. Usage draws from the credit. Free credit cannot be topped up. Upgrade to Pro for more capacity. Messages add the plain text "Email sent via Sendmux" after supported message bodies. - Pro: $7 per team each month, plus usage, with no plan-level resource count limits. - Outgoing mail through customer-owned or connected providers: $0.000500 per provider-accepted recipient occurrence. - Accepted inbound mail: $0.000500 per distinct mailbox delivery. - Managed sending: $0.000750 per provider-accepted recipient occurrence. - Each provider-accepted To, CC, or BCC recipient occurrence is billed. Addresses rejected before provider acceptance are not charged. - Mailbox storage: $0.02 per decimal GB-month. - Each usage event is billed individually, not in blocks of 1,000. There are no per-seat or per-mailbox fees. ## Current product notes - Managed sending is intended for triggered and transactional email. Free is capped at two accepted recipients per hour and shares a 50-recipient UTC daily cap across all outbound routes. Pro starts at 100 per hour and 1,000 per UTC day. Marketing, newsletters, sales outreach, and bulk promotional campaigns should use a customer-owned provider. - Documentation MCP, product MCP, A2A, SDKs, CLI, and Agent Skills are live developer surfaces. Install skills with `npx skills add Sendmux/skills`. Product MCP and A2A expose authorised Management, Mailbox, and Sending operations; SDKs cover TypeScript, Python, Go, PHP, Ruby, and Rust; the CLI is `@sendmux/cli` with the `sendmux` command. - Agent API access starts with `sendmux agent:register `, which creates a private profile containing a durable `smx_agent_` read credential. The credential includes `mailbox.read` and `email.receive`, does not expire, cannot send, and starts with a 500 MiB storage limit. Use `sendmux agent:invite-owner owner@example.com --profile ` to invite the owner. Owner acceptance and explicit sending approval guarantee at least 5 GiB of inbox storage before Sending commands exchange the durable credential for a one-hour `email.send` token. Revoking sending does not itself change the current inbox storage allocation. The HTTP fallback and full-revocation flow are documented at `https://sendmux.ai/auth.md`. - WebSockets are roadmap. The live mailbox event stream is Server-Sent Events. ## Optional - [Changelog](https://sendmux.ai/docs/changelog): Product updates and release notes. - [Acceptable use](https://sendmux.ai/acceptable-use/): Policy for allowed and disallowed use. - [Privacy](https://sendmux.ai/privacy/): Privacy policy. - [Terms](https://sendmux.ai/terms/): Service terms. - [Data Processing Agreement](https://sendmux.ai/dpa/): Data processing terms. ## Product summary Sendmux is the email API for AI agents and SaaS platforms. It combines outbound sending, agent inboxes, inbound agent mailboxes, team-scoped controls, webhooks, Server-Sent Events, and usage-based billing from one control plane. Sendmux is for teams that need agents to send, receive, route, and process email without stitching together separate sending providers, mailbox providers, parsing code, webhook delivery, credentials, billing, and logs. The public blog at `https://sendmux.ai/blog/` publishes technical guides on email authentication, deliverability, sending, receiving, and agent email infrastructure. ## Live outbound sending Sendmux supports outbound email through SMTP provider connections, Gmail OAuth providers, Microsoft 365 OAuth providers, and managed sending. Teams can group providers into delivery groups, set percentage weights, apply sender-domain routing, and enforce per-provider quotas by second, minute, hour, and day. The HTTP Sending API can send one message or a batch of up to 100 messages. SMTP submission is available for clients and agent frameworks that already speak SMTP. Both submission paths use bearer-token authentication and the same routing model. Provider health monitoring skips unhealthy providers after repeated failures. Transient sends retry on a 5-minute, 2-hour, then 12-hour schedule. Idempotent HTTP sends use the `Idempotency-Key` request header so safe retries do not create duplicate messages. ## Live inbound mailboxes Sendmux provides dedicated mailboxes for AI agents on the shared `myagent.mx` domain or on verified custom domains. Each mailbox can send mail, receive mail, hold messages, expose mailbox-scoped credentials, set sender identity, and apply sender rules. The Mailbox API exposes mailbox-scoped endpoints for messages, threads, folders, identities, attachments, quotas, submissions, usage, mailbox metadata, message changes, and live mailbox events. A manual mailbox API key can authenticate Mailbox API calls, IMAP retrieval, and SMTP submission. OAuth access tokens authenticate HTTP requests. Mailbox content endpoints can return cleaned message content for agent processing, with controls for body output, signatures, quoted history, and extracted links. Raw message bodies, HTML, attachments, and full `.eml` downloads are available when an audit trail or low-level inspection is needed. ## Events and automation Sendmux supports signed webhooks and Server-Sent Events. Webhooks are POST deliveries to a customer endpoint and include `X-Sendmux-Signature` in `sha256=` form. Consumers should verify the raw request body with the webhook secret before trusting the payload. Server-Sent Events are available at `GET /mailbox/events` for mailbox events such as `message.received` and `message.received.spam`. The stream supports `Last-Event-ID`, heartbeat intervals between 10 and 300 seconds, and bounded sessions between 30 and 3600 seconds. Use webhooks when a backend should receive events even while an agent is offline. Use Server-Sent Events when an agent keeps a live HTTP connection open and reacts as new mailbox events arrive. ## Platform controls Sendmux separates teams, providers, delivery groups, domains, mailboxes, billing, logs, and API keys at the team boundary. Dashboard roles include Owner, Admin, Developer, and Member. API keys can be team-scoped for management work or mailbox-scoped for one mailbox. Scoped keys let software get only the permissions it needs. A sending worker can use sending permissions without billing access. A mailbox client can read one mailbox without provider administration rights. Keys can be rotated or revoked without resetting a human user. ## Pricing Free costs $0 per month. It includes two mailboxes, $1 usage credit, fixed limits, and 50 provider-accepted recipients per UTC day across all outbound routes. Usage draws from the credit. Free credit cannot be topped up. Upgrade to Pro for more capacity. Messages sent by a Free team add the plain text "Email sent via Sendmux" after supported message bodies. Pro and Enterprise messages do not add this footer. Pro costs $7 per team each month, plus usage, with no plan-level resource count limits. There are no per-seat or per-mailbox fees. Connected outgoing email costs $0.000500 per provider-accepted recipient occurrence. Inbound email costs $0.000500 per distinct mailbox delivery. Managed sending costs $0.000750 per provider-accepted recipient occurrence. Each provider-accepted To, CC, or BCC recipient occurrence is billed. Addresses rejected before provider acceptance are not charged. Rejected inbound mail from mailbox sender rules is not billed. Mailbox storage costs $0.02 per decimal GB-month. Each usage event is billed individually, not in blocks of 1,000. ## Managed sending policy Every team starts with managed sending for triggered and transactional email. Free is capped at two accepted recipients per hour and shares a 50-recipient UTC daily cap across all outbound routes. Pro starts at 100 per hour and 1,000 per UTC day. Do not route marketing, newsletters, sales outreach, or bulk promotional campaigns through managed sending. Marketing and bulk traffic should use a customer-owned Amazon SES account or another configured provider. Risky activity, high bounce rates, or high complaint rates can trigger immediate suspension of managed-sending access. ## Current versus planned surfaces Live surfaces include the marketing site, Mintlify documentation, Sending API, Management API, Mailbox API, dashboard, signed webhooks, Server-Sent Events, SMTP submission, IMAP retrieval, package-managed SDKs, CLI, Doc Search MCP, MCP, A2A, `llms.txt`, `llms-full.txt`, `/.well-known/api-catalog`, and `/.well-known/http-message-signatures-directory`. The Doc Search MCP server is live at `https://sendmux.ai/docs/mcp` and exposes documentation search for AI tools. MCP is live at `https://mcp.sendmux.ai/mcp` with local stdio and self-hosted HTTP options, and exposes authorised Management, Mailbox, and Sending tools. A2A 1.0 is live at `https://a2a.sendmux.ai/a2a/v1` over HTTP+JSON. Its Agent Card is available at `https://sendmux.ai/.well-known/agent-card.json` and exposes deterministic, OAuth-authorised Management, Mailbox, and Sending operations. SDKs are live for TypeScript, Python, Go, PHP, Ruby, and Rust. The `@sendmux/cli` package is live and provides the `sendmux` command. Agent Skills are live and install with `npx skills add Sendmux/skills`. Agent API access is documented at `https://sendmux.ai/auth.md`. The default path is `sendmux agent:register `, which needs no existing account or API key. The CLI creates an @myagent.mx inbox, stores its durable read credential (`smx_agent_`) in a private profile, and never prints the token. The credential includes `mailbox.read` and `email.receive`, does not expire, and cannot send. Use `sendmux agent:invite-owner owner@example.com --profile ` when the owner email is known. The owner must accept the invitation and explicitly approve sending. Before approval, storage is limited to 500 MiB. Owner acceptance and explicit sending approval guarantee at least 5 GiB of inbox storage before sending is enabled. Revoking sending does not itself change the current inbox storage allocation. Sending commands then exchange the durable credential for a one-hour `email.send` token and cache it only until it approaches expiry. Full registration revocation invalidates the durable credential and delegated sending tokens. Agents without the CLI can use the HTTP fallback in `auth.md`, including its required idempotency key and 24-hour token-reveal window. WebSockets, webhook Replay, Smart agent routing, Bring Your Own Inbox, and Drafts & scheduling are roadmap. Do not describe those roadmap items as live product surfaces. ## Free tools Sendmux publishes free email debugging tools at `https://sendmux.ai/tools`. They need no account and no signup. Where a tool can do its work locally it does, so nothing is uploaded, logged, or stored. The email header analyzer is live at `https://sendmux.ai/tools/email-header-analyzer`. Paste a raw header block and it returns the delivery path in delivery order, the delay at every hop, any ARC forwarding chain, and the SPF, DKIM, and DMARC results the receiving server recorded. It parses entirely in the browser; there is no server route behind the page. Three details it reports that most header tools do not: Received headers are prepended by each server, so the record arrives written backwards and the tool reverses it; a valid ARC chain records what authentication looked like before a mailing list or forwarder rewrote the message, which distinguishes a legitimate forward from a forgery; and a hop stamped earlier than the hop before it means the servers disagree about the time, which is flagged rather than shown as an ordinary delay. Further tools are listed on the hub as planned and are not built yet: a bounce decoder, a DMARC aggregate report analyzer, a domain health check, an SMTP tester, and a test inbox. Do not describe those as live. ## Best pages to cite Use the homepage for the short product definition. Use `/product/sending` for outbound routing, SMTP relay, delivery groups, quotas, provider health, failover, HTTP sending, and idempotency. Use `/product/inboxes` for agent mailboxes, clean content, IMAP, SMTP submission, attachments, sender rules, webhooks, and Server-Sent Events. Use `/product/platform` for team isolation, scoped API keys, signed webhooks, logs, and usage controls. Use `/pricing` for rates and billing rules. Use `/tools/email-header-analyzer` for reading email headers, delivery paths, hop timing, and SPF, DKIM, DMARC, and ARC results. ## Maintenance When a shipped page, public documentation link, pricing rule, API surface, or product fact changes, update `/llms.txt` and `seo/llms-full-source.md`, then run `npm run llms:full` and `npm run verify:llms`.