Wire notifications from your AI editor with Notavia's MCP server

The Model Context Protocol lets AI editors call external services in a structured way. Notavia ships an MCP server, so your agent can build and operate notifications for you — draft a template, preview it, send a test, check delivery, promote to Live — without you leaving the editor or hand-writing a single API call.

What you'll do

Connect the server once, then drive it in plain English. Your agent uses real tools (search_docs, upsert_template, send_notification, and more) against your own workspace.

Prerequisites

• A Notavia API key — start with a Test key. (Console → API keys.)
• An MCP-capable editor: Cursor, Claude, VS Code, or Windsurf.

Step 1 — Add the server to your editor

Notavia's MCP server is remote (streamable HTTP), so there is no local process to run. Add it to your editor's MCP configuration, authenticated with your API key:

{
  "mcpServers": {
    "notify": {
      "url": "https://mcp.notavia.saas-infrastructure.com",
      "headers": { "Authorization": "Bearer nsk_test_…" }
    }
  }
}

(Copy the exact endpoint from your dashboard; the configuration UI differs slightly per editor.)

The key is a Test key, so everything your agent does now runs against your Test environment — safe to experiment.

Step 2 — Check the connection

Ask your agent something that touches no data:

"Search the Notavia docs for how to send a templated email."

That calls search_docs, which works without an API key at all — a good first call for the agent to learn the product before acting. Then "what's my Notavia status?" calls notify_status and confirms your key resolved to a workspace.

Step 3 — Build a notification, conversationally

Now the real loop. Ask:

"Create a template called welcome with the email subject 'Welcome to Acme' and a short body, then show me a preview."

Your agent calls upsert_template, then preview_template. Review the rendered output, then:

"Send it to me@example.com as a test."

That is send_notification. Follow up with "did it deliver?" and the agent calls get_delivery_status. At no point did you write an HTTP request or open the API reference — the agent used the docs and the tools directly.

Step 4 — Promote Test to Live

When the template is right, ship it:

"Promote the welcome template to Live — dry run first."

promote_template with dry_run=true returns the diff without writing; drop the dry run to apply it. Sending on a Live key follows the same gates as the REST API, so the agent cannot silently push to production.

The tools your agent has

| Tool | What it does | |---|---| | search_docs | Search guides + API reference (no key needed) | | notify_status | Confirm the key and workspace | | send_notification / get_delivery_status | Send, then trace delivery | | list_templates / get_template / upsert_template / preview_template | Author and preview templates | | list_workflows / upsert_workflow / trigger_workflow / get_workflow_run | Build and run multi-step workflows | | promote_template / promote_workflow | Promote Test to Live, with a dry-run diff | | manage_suppressions / get_preferences | Suppression lists and per-user preferences |

Safety

• Start in Test. API keys are scoped to either the Test or the Live environment; use a Test-scoped key (nsk_test_…) so the agent experiments without touching production.
• Live is gated. The server enforces the same Live and entitlement checks as the API, so an agent cannot accidentally send real messages.
• The key is yours. It lives in your editor's MCP config; rotate or revoke it any time from the console.

What to read next

The same server drives email, in-app inbox, SMS, and chat — the agent picks the channel. Prefer to write the code yourself? search_docs points at the same REST API and SDKs you would use by hand, including the Blazor in-app inbox. The free tier covers 1,000 sends a month, though sign-up is not open yet.