MCP tools reference

What does each tool do?

Reading: list_numbers returns the connected numbers with status, quality rating and messaging tier, and is where every other call starts. get_number_health returns a live Meta snapshot plus the recent monitoring timeline, webhook events and poller checks together. get_api_credentials returns the Graph API base URL, WABA id, phone number id and, where available, the account's own access token.

Sending: send_message sends free text inside the 24 hour window, where replies are free. send_template sends an approved template, which is how a conversation is started with someone who has not written. list_templates, create_template and get_template_status cover the template lifecycle.

Wiring: set_webhook_destination routes incoming messages and app echoes to your system, and passing null resets it. get_connect_link returns the dashboard link and steps for connecting a new number. get_docs returns the product documentation as markdown.

Why does connection_id matter?

Because it is what ties an instruction to a specific number, and an assistant cannot guess it.

Every tool that acts on a number takes a connection_id, and list_numbers is the only place one comes from. In practice this means an assistant asked to send a message will call list_numbers first, which is why the first request in a session is often slower than the rest.

It also means an account with several numbers is unambiguous by construction. There is no default number and no implicit selection, so an instruction that does not identify which number is asked about rather than guessed at.

What happens when billing lapses?

The four tools that act on the outside world are gated: send_message, send_template, create_template and set_webhook_destination. Everything that reads stays available.

This is deliberate rather than an oversight. An account that cannot see why it is blocked, or cannot check whether its numbers are healthy, has no route back except a support ticket. Those same reads are also what the health monitoring depends on, so gating them would switch off the monitoring at the moment somebody most needs it.

The practical effect is that a lapsed account still answers questions about its own state and is still monitored. What stops is anything that reaches a customer.

Common mistakes

• Expecting the connector to receive messages. MCP is request and response; incoming traffic needs a webhook.
• Hardcoding a connection_id. It comes from list_numbers and is not a value to write down.
• Assuming a lapsed account is fully offline. Reads and monitoring continue on purpose.

Frequently asked questions

Keep reading