Let your agent build email.
Keep the boundary.

Connect a coding agent to NoticeAPI over hosted MCP or local stdio. It discovers the email tools, acts only where its grant allows, and proves the path in the simulator before real mail.

  • OAuth or bearer key
  • Project-scoped access
  • Deterministic testing
Hosted MCP
mcp client config
{
  "mcpServers": {
    "noticeapi": {
      "type": "http",
      "url": "https://mcp.noticeapi.com/mcp",
      "headers": { "Authorization": "Bearer ntc_…" }
    }
  }
}
Streamable HTTPOAuth or scoped key

Agent-ready by design

One clear prompt. A real implementation path.

MCP exposes more than a send call. The agent can discover the project, prepare a domain, create a template, test an outcome, and follow the recipient event trail with the same tool contract throughout.

Read the agent guide
ClaudeCursorWindsurfGitHub Copilot
Prompt

Add production email with NoticeAPI. Use MCP or OpenAPI. Test in the sandbox.

  1. 01
    DiscoverProjects, domains, templates, sends, and events.
  2. 02
    TestRun the whole path against deterministic recipients.
  3. 03
    ShipMove to the verified project when the result is right.

Bounded authority

The agent can only reach what the grant already can.

Bind access to a project, chosen domains, and the capabilities the job needs. Sending, scheduling, and outcome reads can stay available while broader workspace operations remain out of reach.

  • Project and domain grants are enforced on every tool call
  • Quotas, suppressions, and deliverability pauses still apply
  • Side effects and destructive tools are machine-readable
Explore scoped keys
Key grantproject_123
Fullall capabilities

Every allowed capability inside one project.

Sendingemail:send

Send, schedule, and read recipient outcomes.

Customselected scopes

Choose capabilities, projects, and domains by hand.

Authority can narrow, never widen.Any key created by this grant is equal to or narrower than its parent.

One tool catalog

From first project to recipient outcome.

Give the agent only the job it needs today. Add capabilities later without changing the connection or teaching it another provider.

See every MCP tool
CapabilityScopesWhat the agent can do
Projects and accessprojects:* · api-keys:write

List and operate projects, then issue or revoke grants no broader than the caller.

Transactional emailemail:send

Send, batch, schedule, cancel, reschedule, and inspect recipient outcomes.

Domainsdomains:read · domains:write

Create, inspect, verify, and remove sender domains.

Templatestemplates:write

Create, inspect, update, and remove stored templates.

Marketingaudiences · broadcasts · automations

Operate opted-in contacts, broadcasts, and lifecycle sequences.

Receivingreceiving:read · receiving:write

Manage API receiving domains and read inbound messages.

Operationswebhooks · suppressions

Manage signed event endpoints and recipient suppression state.

Test first

Prove the whole flow without mailing a real person.

The simulator exercises the send, status, webhook, and suppression pipeline with deterministic recipients. Your agent can see the result and fix the integration before it touches a verified sender.

  1. 01Send from the sandbox[email protected]No sender reputation is involved.
  2. 02Choose the outcome[email protected]Delivered, bounced, complained, or suppressed.
  3. 03Exercise the real event pathThe simulator fires the same signed x-noticeapi-* events.
  4. 04Read the recipient timelineThe agent fetches the result and can correct the workflow.

Ready for an agent?

Give it the tools. Keep the boundary.

Create a project-scoped key, connect MCP, and let the agent prove the email path in the simulator first.