Model Context Protocol
Taverity MCP Server
Connect Taverity to Claude, ChatGPT, or any AI assistant that speaks the Model Context Protocol, and let it draft and queue personal follow-up texts for you. Every message still lands on your iPhone and sends from your own number, only after you approve it. The assistant queues; you send.
Connection details
https://portal.taverity.com/mcp- Transport
- Streamable HTTP (remote MCP). No install — connect by URL.
- Authentication
- OAuth 2.1 with PKCE and Dynamic Client Registration — sign in with your Taverity account, no manual token to paste.
- Scopes
mcp:readto read your account and preview,mcp:sendto queue a send.- Requirements
- A Taverity account and the Taverity iPhone app, which sends the queued texts from your own number.
Connecting it
In any MCP-capable assistant — Claude, ChatGPT, or another client — add Taverity as a custom or remote MCP connector using the server URL above, then complete the sign-in. There's nothing to install and no token to copy: you authenticate with your Taverity account through your browser and approve the read and send access the assistant should have.
Once connected, ask your assistant to draft a follow-up — for example, "text my Past Clients list a spring check-in." It previews what would be sent, you confirm, and it queues the message to your iPhone for approval. Each client's exact menu differs and evolves, so follow your assistant's own steps for adding a connector.
Tools
whoamimcp:readReturns the connected account, granted scopes, and the current date in your time zone — so the assistant resolves "next Tuesday" against your calendar, not its own. No side effects.
get_contact_listsmcp:readLists your named contact lists (iPhone Contacts groups) with member counts, so the assistant can resolve "my Past Clients list" to the exact stored name. Returns names and counts only — never phone numbers.
preview_one_time_messagemcp:readValidates a draft and returns exactly what would be sent, plus any warnings — without sending anything. The assistant shows you this before it sends so you can confirm the wording and recipients.
send_one_time_messagemcp:sendQueues the message to your Taverity account. It still lands on your iPhone for approval before it goes out — nothing sends from a server. Requires the separate mcp:send scope, so a read-only connection can never send.
How sending stays safe
Sends from your own iPhone
The assistant only queues a message. It surfaces on your iPhone in the Taverity app and sends from your own number after you approve — there is no server-side sending and no shared shortcode.
Nothing sends without approval
Every queued message waits for your tap in the app by default. The assistant is prompted to preview and get your confirmation before it ever calls the send tool.
Your contacts stay private
List members are resolved on our server, not in the model. Phone numbers are stripped from every tool response, so your contacts' numbers never enter the assistant's context.
Scoped, revocable access
You grant read and send as separate scopes during sign-in, and can revoke the connection any time from the Taverity portal. Access tokens are short-lived and rate-limited.
Getting started
- Download the Taverity iPhone app and create your account.
- Add the MCP server URL to your AI assistant and sign in.
- Ask your assistant to draft a follow-up — approve it on your phone, and it sends from your own number.
Frequently asked
Can the assistant send a text without me seeing it?
No. The send tool only queues a message; it surfaces on your iPhone in the Taverity app and sends from your own number after you approve. The assistant is also prompted to preview and get your confirmation before it calls send at all.
Does the assistant see my contacts' phone numbers?
No. When you target a named list, member phone numbers are resolved on our server and stripped from every tool response. The assistant only ever sees list names and counts, not numbers.
How do I disconnect it?
Remove the connector in your LLM client, and revoke the connection from the Taverity portal. Access tokens are short-lived and every MCP request is rate-limited per token.
Is this the same as the API?
No. The MCP server is for LLM assistants and is available to any Taverity user. The REST API is a team-plan trigger surface for your own backends and low-code tools. Both route messages through your iPhone for approval.