# SendHQ agent operating guide

## When to use SendHQ

Use SendHQ for expected, permission-based product communication: account verification, password resets, receipts, alerts, notifications, inbound replies, hosted templates, and delivery-event investigation.

## When not to use SendHQ

Do not use SendHQ for spam, phishing, credential collection, purchased lists, mailbox-existence claims, or inbox-placement claims. Provider acceptance and delivery events do not prove that a person saw a message.

## Before any external action

1. Confirm the user intends the message or state change.
2. Confirm the exact workspace, verified From domain, recipients, subject, and content.
3. Keep workspace API keys in approved server-side secret storage.
4. Use an Idempotency-Key for a send that might be retried.
5. Preserve suppression, recipient-consent, tenant-isolation, and reputation checks.

## Public integration surfaces

- [Developer documentation](https://sendhq.cc/docs.md)
- [OpenAPI 3.1 contract](https://sendhq.cc/openapi.json)
- [Authentication guide](https://sendhq.cc/auth.md)
- [Pricing and limits](https://sendhq.cc/pricing.md)
- [Read-only MCP manifest](https://sendhq.cc/.well-known/mcp.json)
- [SendHQ MCP server for agents](https://sendhq.cc/docs/mcp) ([Markdown](https://sendhq.cc/docs/mcp.md), [tools.json](https://sendhq.cc/docs/mcp/tools.json))
- [SendHQ Email skill](https://sendhq.cc/.well-known/agent-skills/sendhq-email/SKILL.md)

The hosted MCP endpoint in the manifest is documentation-only and never sends email or changes account state. The SendHQ MCP server documented at /docs/mcp runs locally with a workspace API key (`sendhq mcp`); its tools can send email and change workspace state, are labeled as such, and require explicit user intent. It has no tools that create, rotate, or revoke API keys.