Termii MCP

A hosted Model Context Protocol (MCP) server that gives AI assistants like Claude and ChatGPT the ability to send SMS and WhatsApp messages, run campaigns, and manage phonebooks through Termii, in natural language.

It wraps the zeevx/lara-termii package and exposes the full Termii API as 22 MCP tools, built with the official laravel/mcp package.

Live at termiimcp.com. The landing page turns your Termii API key into a personal connector link that you add to your assistant.

How it works

The MCP server is multi-tenant and stateless. Each user connects with a server URL that carries an encrypted token in a key query parameter:

https://termiimcp.com/mcp?key=<ENCRYPTED_TOKEN>

This is because Claude's and ChatGPT's connector forms only let you edit the URL; Claude has no field for a static bearer token (it supports OAuth only). To avoid putting the raw key in the URL, the landing page posts your key to /connect, which encrypts it with the server's APP_KEY and returns the token. Only this server can decrypt it, and every tool call decrypts the token to reach the underlying Termii account. Programmatic clients may instead send the raw key as an Authorization: Bearer <key> header.

The key is never stored; it lives only for the duration of the request. There is no database, no sign-up, and no shared credentials.

Related MCP server: WAHA MCP

Capabilities

Read-only tools are annotated IsReadOnly, deletes IsDestructive, and updates IsIdempotent, so clients can surface the right confirmations.

Requirements

• PHP 8.4+

• Composer

• Node.js 20+ (for building the landing page assets)

Installation

git clone https://github.com/<your-org>/termiimcp.git
cd termiimcp

composer install
npm install

cp .env.example .env
php artisan key:generate

npm run build

Then serve it:

php artisan serve
# MCP endpoint: http://localhost:8000/mcp
# Landing page: http://localhost:8000

Or run the full dev stack (server + Vite) with composer dev.

Configuration

This is a light, database-free application. Cache, sessions, and the queue use file/sync drivers out of the box, so there is nothing to migrate.

Because every request authenticates with the caller's own key, no TERMII_API_KEY is required on the server. The TERMII_* values in .env act only as optional fallbacks for a single-tenant deployment. In production, set APP_URL to your public HTTPS domain so generated URLs (and the OG image) resolve correctly.

Connector tokens are encrypted with the app's APP_KEY, so keep it stable in production. Rotating APP_KEY invalidates every existing connector link, and users would need to reconnect.

The public endpoint is rate limited per API token (60 requests/minute), configured in app/Providers/AppServiceProvider.php.

Connecting from an AI client

The easiest path is the termiimcp.com landing page: paste your Termii API key and it generates your personal connector link (with the encrypted token) plus one-click buttons.

Under the hood, your link looks like https://termiimcp.com/mcp?key=<ENCRYPTED_TOKEN>, where the token comes from the /connect endpoint.

Claude: the "Add to Claude" button opens a prefilled connector dialog of the form:

https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=Termii+MCP&connectorUrl=<URL-ENCODED_CONNECTOR_URL>

ChatGPT: enable Settings, Connectors, Advanced, Developer mode, choose Create, and paste the connector URL. (ChatGPT connectors require a paid plan.)

Once connected, ask things like "check my Termii balance" or "send an SMS to 2348012345678".

Testing

php artisan test --compact

Tests use Laravel's HTTP client fakes, so they never hit the real Termii API.

Project layout

app/Mcp/Servers/TermiiServer.php                  The MCP server: registers tools + instructions
app/Mcp/Tools/                                    One class per Termii action (22 tools)
app/Mcp/Tools/Concerns/InteractsWithTermii.php    Resolves the client from the bearer token
app/Http/Controllers/HomeController.php            The landing page
routes/ai.php                                     Registers the web MCP server at /mcp

Note: LaraTermii::addContactsFromFile() is intentionally not exposed as a tool. It reads a file from a server-side storage disk, which a remote AI client cannot reference. Use add-contacts-from-contents (raw CSV text) instead.

Contributing

Contributions are welcome. Please run vendor/bin/pint and php artisan test before opening a pull request. New tools follow the pattern in app/Mcp/Tools and register in TermiiServer::$tools.

Credits

• Laravel MCP

• zeevx/lara-termii

• Termii

License

Released under the MIT License.