MCP server

What it is

@elaanio/mcp is an MCP server that puts the Elaan management API in front of an AI assistant. Once it is connected, you can ask for the work instead of writing the calls: create a notification type, add a template for a brand, trigger a test send, or find out why yesterday's email never arrived.

It speaks stdio, so it runs locally as a subprocess of your assistant. Nothing is hosted by us, and no traffic goes anywhere except from your machine to the Elaan API.

Install

You need a service key (sk_…) from the console. There is nothing to install ahead of time; npx fetches the package on first run.

claude mcp add elaan --env ELAAN_API_KEY=sk_… -- npx -y @elaanio/mcp

{
  "mcpServers": {
    "elaan": {
      "command": "npx",
      "args": ["-y", "@elaanio/mcp"],
      "env": { "ELAAN_API_KEY": "sk_…" }
    }
  }
}

{
  "mcpServers": {
    "elaan": {
      "command": "npx",
      "args": ["-y", "@elaanio/mcp"],
      "env": { "ELAAN_API_KEY": "sk_…" }
    }
  }
}

{
  "servers": {
    "elaan": {
      "command": "npx",
      "args": ["-y", "@elaanio/mcp"],
      "env": { "ELAAN_API_KEY": "sk_…" }
    }
  }
}

{
  "mcpServers": {
    "elaan": {
      "command": "npx",
      "args": ["-y", "@elaanio/mcp"],
      "env": { "ELAAN_API_KEY": "sk_…" }
    }
  }
}

{
  "mcpServers": {
    "elaan": {
      "command": "npx",
      "args": ["-y", "@elaanio/mcp"],
      "env": { "ELAAN_API_KEY": "sk_…" }
    }
  }
}

The server refuses to start without a key, and it tells you so rather than failing on the first call. If your config passes an unexpanded placeholder such as ${ELAAN_API_KEY}, that is treated as missing too, because a literal dollar sign in an auth header is never what anyone meant.

Where the key lives

The API key is read from the environment and is never a tool argument. That is deliberate. A key passed as an argument becomes part of the conversation: it lands in transcripts, in scrollback, and in whatever logging sits between the assistant and you. Keeping it in the process environment means the model never sees it and cannot repeat it back.

The tools

Twenty-six of them, covering the management surface. You do not need to learn the names; the assistant picks them. They are listed here so you can see the shape of what it can reach.

Notification types

list_notification_types, get_notification_type, create_notification_type, update_notification_type, delete_notification_type, restore_notification_type

Types are soft deleted, which is why restore exists. Other objects reference a type by key, so the record has to survive.

Templates

list_templates, get_template, create_template, update_template, delete_template, list_templates_affected_by, preview_email_render

list_templates_affected_by answers "if I rename this variable, what breaks", and preview_email_render renders a template against a brand and a set of variables without sending anything. Both are far more useful in a conversation than in a dashboard.

Brands

list_brandings, create_branding, update_branding, delete_branding

The tenant default brand cannot be deleted, and the server returns the same error the API does rather than pretending otherwise.

Contacts and preferences

list_contacts, get_contact, create_contact, get_contact_preferences, set_contact_preference

Sending and debugging

trigger_notification, get_notification_event, list_deliveries, get_delivery_stats

This is the group that earns its keep. Trigger a send, take the returned event id, and ask what happened to it per channel, all without leaving the conversation.

What it deliberately cannot do

Transport credentials are absent, not gated. There are no tools for /v1/email-transport or /v1/push-transport, so the server cannot read or write your SMTP, SES, SendGrid, Mailgun, Expo or FCM credentials.

That is a deliberate scope decision rather than a permission check. A tool that exists but refuses is still a tool an assistant will try, describe and reason about. One that was never registered cannot be called by mistake, cannot be talked into running, and cannot leak a secret through an error message. Configure transports in the console.

Things worth asking it

• "Create a notification type for order shipped, with email and in-app on by default."
• "Add an email template for it, and show me how it renders for the Northwind brand."
• "Trigger it for contact crm-12345 and tell me what happened on each channel."
• "Which templates use the order_id variable?"
• "What is the delivery success rate this week, and which channel is worst?"

The last two are the ones people are surprised by. Both are questions about your configuration rather than a single record, and both are tedious to answer by clicking.