Email your AI agent operates and you supervise

Agent quickstart

For an AI agent reading this page: how to find AISend, get access, make the first send and know what to do when it fails. Everything below is what the service does today.

1. Discover

• Plain-text summary: https://aisend.app/llms.txt (full: https://aisend.app/llms-full.txt)
• Markdown pages: https://aisend.app/mcp.md, https://aisend.app/pricing.md, https://aisend.app/index.md
• MCP server card: https://aisend.app/.well-known/mcp/server-card.json
• Official MCP Registry: app.aisend/aisend at registry.modelcontextprotocol.io

2. Authenticate

• OAuth sign-in by the human owner (recommended for assistants). Connect to https://api.aisend.app/mcp. An unauthenticated request gets HTTP 401 with a WWW-Authenticate header pointing at the protected-resource metadata, so an OAuth-capable client discovers the authorization server, registers itself (RFC 7591) and runs the authorization-code flow with PKCE S256. The owner signs in on aisend.app and presses Approve. The result is an API key named "MCP OAuth: <client>" that works only on the MCP endpoint, expires after 90 days (reconnect after that), is limited to 100 emails a day and can be revoked on the API Keys page.
• API key as a Bearer header. Send Authorization: Bearer re_... to the hosted server or the REST API (https://api.aisend.app/api/v1/...). A key the owner created has the account's full API access and does not expire until revoked.
• Self-serve agent signup (local server or REST). POST https://api.aisend.app/api/v1/auth/agent-signup with {"name": "<agent name>"} (the local MCP tool agent_signup does the same) returns an API key without a browser. The account cannot send yet: a human must subscribe to a plan and the account needs a verified sending domain. The hosted server does not offer this tool.

3. Minimal call sequence

Tool names and arguments are the real ones; ids and addresses are examples.

list_domains({})

validate_email({
  "email": "customer@example.com"
})

send_email({
  "from": "Acme <hello@acme.com>",
  "to": "customer@example.com",
  "subject": "Your receipt",
  "html": "<p>Thanks for your order.</p>",
  "idempotency_key": "order-1042-receipt"
})

get_email_status({
  "email_id": "<id from step 3>"
})

list_emails({
  "limit": 10
})

4. Limits and quotas

• Recipients per send: up to 50.
• Monthly quota by plan: Pro 50,000 emails/month; Business 100,000 emails/month; Scale 500,000 emails/month (daily cap 5,000, about 150,000 a month today).
• Daily cap by plan (UTC day): Pro 2,500, Business 5,000, Scale 5,000. New accounts ramp up: 50 a day in the first 24 hours, 200 until day 3, 500 until day 7, 1,000 until day 14, then the plan cap. A newly verified domain has its own lower cap for its first week.
• An OAuth-created connection is limited to 100 emails a day, whatever the plan.
• Request rate per API key: 50 requests/second on Pro, 100 on Business and Scale, 10 when the account cannot send. Over the limit: HTTP 429 with Retry-After: 1 and X-RateLimit-Limit / X-RateLimit-Remaining headers.
• Monthly validation quota by plan: Pro 5,000, Business 25,000, Scale 100,000.

5. Idempotency

• idempotency_key is required on send_email (up to 255 characters; sent as the Idempotency-Key header).
• Retrying with the same key and the same request returns the first result and sends nothing new.
• Reusing a key for a different request fails with HTTP 409, code IDEMPOTENCY_KEY_REUSED. Use a new key for a different email.
• Use a key derived from the business event (for example order-1042-receipt) so a restarted agent does not send a second copy.

6. Errors the agent must handle

Over MCP a failed call returns an error result whose text is "AISend API request failed (<status>): <message>". The machine-readable code field and upgradeUrl are returned by the REST API; the MCP text carries the status and the message, so match on both. The ChatGPT connector endpoint replaces plan and billing messages with a neutral note that the account cannot do that right now because of its limits.

7. Hand off to the human

• Before the first send in a conversation, show the human the sender, recipients, subject and body and wait for a clear yes (most clients also prompt on send_email because it is marked destructive).
• After sending, report the email id and the status exactly as returned, and tell the human where to review it: the Emails page of the AISend dashboard.
• For 402, 403 domain or email verification, and 429 limits: stop and give the human the exact message; these are decisions or actions only the account owner can take.
• Never ask the human to paste an API key into the chat; use the OAuth sign-in or an environment variable.

Connect in 60 seconds

With OAuth, the client opens AISend in your browser. You sign in and see a consent screen that names the client and your account and lists what the assistant will be able to do: send email from your verified domains under your normal limits, read sent emails and delivery status, validate addresses (uses your validation quota), and list your sending domains and API key names. Press Approve (or Deny) and you return to the client. The connection is an API key named MCP OAuth: <client> that you can revoke on the API Keys page.

ChatGPT

1. Open ChatGPT settings and enable developer mode / custom connectors (availability depends on your plan and workspace).
2. Add a custom MCP connector with the URL https://api.aisend.app/mcp (or https://api.aisend.app/mcp/chatgpt, the same tools with plan and billing messages replaced by a neutral note) and OAuth authentication.
3. Sign in to AISend in the window that opens and press Approve.
4. Ask: “List my last 5 AISend emails and their delivery status.”

Claude (claude.ai)

Settings, Connectors, add a custom connector with https://api.aisend.app/mcp, then sign in and approve.

Claude Code

Hosted server with OAuth (then run /mcp and authenticate):

claude mcp add --transport http aisend https://api.aisend.app/mcp

Or the plugin:

/plugin marketplace add airdropfi/aisend-claude-plugin
/plugin install aisend@aisend-plugins

Local server with an API key (also works in Claude Desktop):

claude mcp add aisend -e AISEND_API_KEY=re_your_api_key -- npx -y aisend-mcp

Cursor

Cursor's own redirect is not accepted by AISend's OAuth sign-in, so Cursor uses an API key as a Bearer header. Create one on the API Keys page and add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "aisend": {
      "url": "https://api.aisend.app/mcp",
      "headers": { "Authorization": "Bearer ${env:AISEND_API_KEY}" }
    }
  }
}

Or the local server: "command": "npx", "args": ["-y", "aisend-mcp"] with AISEND_API_KEY in env.

Gemini CLI

gemini extensions install https://github.com/airdropfi/aisend-gemini-extension

Restart Gemini CLI, run /mcp auth aisend, sign in in the browser and press Approve.

VS Code (GitHub Copilot)

1. Run Chat: Open Customizations and select MCP Servers.
2. Select Add Server, choose HTTP, and enter https://api.aisend.app/mcp.
3. Give it the ID aisend, choose where to store it, and sign in when prompted.

Cline and other MCP clients

Cline, in cline_mcp_settings.json:

{
  "mcpServers": {
    "aisend": {
      "type": "streamableHttp",
      "url": "https://api.aisend.app/mcp",
      "headers": { "Authorization": "Bearer re_your_api_key" }
    }
  }
}

Any client that runs a local command can use {"command": "npx", "args": ["-y", "aisend-mcp"], "env": {"AISEND_API_KEY": "re_your_api_key"}}. Clients that read the Official MCP Registry find AISend as app.aisend/aisend. We have not tested every client; if yours takes a URL, use the hosted endpoint above with OAuth or a Bearer key.