Email API for AI builders and coding agents.

Connect a project through hosted MCP. Let your coding agent build an email integration and test it with the simulator. Inspect the result before production sending.

Start building Connect MCP

Example permission grant

Task

Add a receipt email to the checkout application.

Project: checkout-app
Domain: mail.example.com
Capabilities: email:send · templates:write
Discovery: Hosted MCP · OpenAPI 3.1

Outside the key’s scopeBilling · members · other projects

Coding agent interfaceHosted MCP

Always-running agentSend + inbound events

Machine contractOpenAPI 3.1

Key boundaryCapability · project · domain

Three parts your team can see.

The human chooses the boundary and reviews the implementation. The agent gets a real contract instead of guessing endpoints.

Choose where the agent can work

Connect with interactive MCP authorization or issue a key limited to the project, capabilities, and sender domains the job needs.

Project + capability + domain scope

Let the agent read the real contract

MCP exposes structured tools. OpenAPI 3.1 describes the same REST resources, inputs, responses, and stable error codes.

Hosted MCP + OpenAPI 3.1

Run a simulator test

The agent implements the application path, sends to a simulator recipient, and inspects the delivery record. Verify your sender before production.

Code in your app + delivery record

An API key allows specific capabilities, not unlimited access.

NoticeAPI evaluates the requested capability, project, and domain before the normal delivery and policy controls. A permission failure is a stop condition, not something the agent should work around.

Read the scoped-key guide

Capability

email:send · templates:write

The credential must carry the operation before the request reaches the resource.

Project

checkout-app

Project-local email, domains, templates, logs, webhooks, and suppressions stay in this boundary.

Sender domain

mail.example.com

A selected API key can use only the domains it names, and production senders must still verify.

Platform safety

always enforced

Suppression, unsubscribe, quota, sender, workspace, and acceptable-use checks still run on every request.

Stable stop codesproject_not_alloweddomain_not_allowedinsufficient_scoperecipient_suppressed

Use the interface your agent host supports.

MCP gives a coding agent a tool interface. An always-running agent should use REST, Node, or SMTP from your own worker and handle inbound events in that runtime.

Hosted MCPhttps://mcp.noticeapi.com/mcpStreamable HTTP with interactive OAuth or an explicit bearer key. Choose capabilities and project access during authorization.Connect hosted MCP Local MCP previewnoticeapi mcpLocal stdio is available in the unreleased SDK 0.2.0 source. Use hosted MCP until the npm release is available.Read local setup OpenAPI 3.1/openapi.jsonGenerate or inspect direct REST requests from the canonical machine-readable contract used by the platform.Open the schema Node SDK and CLInpm install noticeapiUse typed application calls, project selection, and direct commands when MCP is not the right runtime interface.Read SDK setup

Send, receive, and act from your application.

Keep the coding agent inside one project while it builds the workflow. Keep a production agent in your own runtime when it must make decisions over time.

Transactional emailAdd receipts, resets, verification, and alerts with retry-safe sends.Marketing emailCreate opted-in audiences, broadcasts, schedules, and recipient results.Lifecycle automationsConnect audience triggers to stored template steps, delays, and stops.Inbound emailRoute signed inbound events to your worker, then fetch parsed messages and attachments on a paid workspace.

Clear boundaries for humans and agents.

Does a coding agent need full workspace access?

No. Use sending-only access for a runtime integration or a custom API key limited to selected capabilities, projects, and domains. Full access is available for a trusted personal agent, but it is not the default requirement.

Can an agent create another API key?

Only when its own API key includes api-keys:write. Any credential it creates must be equal to or narrower than the credential making the request.

Can MCP bypass suppressions or plan limits?

No. The API key decides which operation may be attempted. Project boundaries, domain rules, suppressions, unsubscribe state, quotas, pauses, and acceptable-use controls are still enforced afterward.

Do we have to use MCP in production?

No. MCP can help the coding agent discover and implement the integration. Your application can run the finished workflow through REST, the Node SDK, or SMTP.

Does NoticeAPI run the AI agent?

No. MCP gives a coding client a scoped tool connection. For an always-running agent, run your own model and worker, then use NoticeAPI to send email and receive signed inbound events.

Connect one project and choose the right agent interface.

Start with a scoped key. Use MCP in a compatible coding client, or give your own worker direct API access for an always-running agent.

Start building Connect MCP