Vaartix MCP server
Official# Vaartix MCP server
Let Claude, Cursor and other AI tools send and look up transactional email through
[Vaartix](https://vaartix.com), the email API built in India.
```
"Send the login code 482913 to priya@example.com from noreply@notify.acme.in"
"Did the order confirmation to ravi@example.com get delivered?"
"Show me emails that bounced in the last 7 days"
```
## Tools
| Tool | What it does | API key |
|---|---|---|
| `send_email` | Send one email (HTML and/or text, cc, bcc, reply-to, tags, scheduled sending, idempotency key) | Send-only or full access |
| `send_batch` | Send up to 20 emails in one call | Send-only or full access |
| `get_email` | One email's details and event timeline; the body only with `includeContent` | Full access |
| `list_emails` | Recent emails, filtered by status, search text or days | Full access |
| `cancel_email` | Cancel a scheduled email | Full access |
| `list_domains` | Your sending domains, with verification status and DNS records | Full access |
| `get_domain` | One domain: whether it's verified and the DNS records it needs | Full access |
Sending tools deliver **real email**. Most MCP clients ask you to approve each tool call, so check
the recipients and content before you approve.
Limits on every call: at most 50 recipients per email (to, cc and bcc together), 20 emails per
batch, and no attachments or custom headers. Email bodies come back only when asked for
(`includeContent`), marked as data from the email, never instructions.
## Setup
You need a Vaartix API key from [app.vaartix.com](https://app.vaartix.com) → API keys. Create a
key just for your AI tool. When asked "Is this key for an AI agent or MCP server?", choose
**Yes, an AI agent**. An AI agent key starts with safe limits:
- test addresses only, until you add the recipients it may email
- 20 emails an hour and 50 a day
- email bodies are never returned (read content is off)
Change these at any time with **Limits** on the API keys page. Widening a limit asks for a fresh
confirmation code; narrowing it doesn't. Limits are enforced by the Vaartix API itself, so they hold
whatever the AI tool asks for.
For sending only, make it a **send-only key limited to your domain**. Use a full-access key only if
you also want the AI tool to read your email logs. Don't reuse your app's production key.
### Claude Desktop
Settings → Developer → Edit Config, then add:
```json
{
"mcpServers": {
"vaartix": {
"command": "npx",
"args": ["-y", "vaartix-mcp"],
"env": { "VAARTIX_API_KEY": "vx_your_api_key" }
}
}
}
```
### Claude Code
```bash
claude mcp add vaartix --env VAARTIX_API_KEY=vx_your_api_key -- npx -y vaartix-mcp
```
### Cursor
`~/.cursor/mcp.json` (or `.cursor/mcp.json` in a project):
```json
{
"mcpServers": {
"vaartix": {
"command": "npx",
"args": ["-y", "vaartix-mcp"],
"env": { "VAARTIX_API_KEY": "vx_your_api_key" }
}
}
}
```
### Other clients
Any MCP client that runs stdio servers works: run `npx -y vaartix-mcp` with `VAARTIX_API_KEY` set.
It needs Node.js 20 or later.
## Configuration
| Variable | Required | Default |
|---|---|---|
| `VAARTIX_API_KEY` | Yes | |
| `VAARTIX_BASE_URL` | No | `https://api.vaartix.com` |
Keep the key out of shared or committed config files. Anyone who can read the config can send
email as you.
## Development
```bash
npm install
npm test # tool tests against a fake API
npm run lint
npm run typecheck
npm run build # outputs dist/index.js (the vaartix-mcp command)
```
## Using the tool definitions
The tools are defined once in `vaartix-mcp/tools` (name, input schema, annotations, the permission
each needs and the API routes it calls), so other MCP servers can offer exactly the same tools:
```ts
import { McpServer } from '@modelcontextprotocol/server';
import { registerTools, TOOLS } from 'vaartix-mcp/tools';
const server = new McpServer({ name: 'vaartix', version: '1.0.0' });
registerTools(server, async (method, path, options) => {
// Call the Vaartix API and return the parsed JSON.
});
```
## License
MIT. Vaartix is a product of NAP Designs, Bengaluru.
TDQS
Scored across 7 tools
Each tool targets a distinct resource and action: send_email vs send_batch differ by batch size, get_email vs list_emails differ by single/collection, and list_domains vs get_domain are clearly scoped. cancel_email is clearly separate from the send tools. No overlaps that would cause misselection.
Consistent verb_noun pattern throughout (cancel_email, send_email, get_email, list_emails, get_domain). Minor deviation: send_batch uses a collective noun rather than matching the singular/plural email convention, and singular vs plural nouns are mixed across get/list pairs.
Seven tools is well-scoped for a transactional email service, covering send, cancel, read single, read list, and domain inspection. Every tool earns its place with no redundancy.
Core workflows (send single/batch, cancel scheduled, inspect emails, inspect domains) are covered. Gaps: no tool to schedule an email even though cancel_email implies scheduling exists, and no domain creation/verification operation, so agents may hit dead ends for setup tasks.