Connectors and MCP

Definition

A connector is a bridge between an AI assistant and another service: your email, your documents, your design tool, your hosting provider… Once the connector is enabled, the assistant can read information in that service (a file, a ticket, a mockup) and sometimes act on your behalf (create a document, publish a page, open a change request).

The MCP protocol

MCP (Model Context Protocol) is an open standard launched by Anthropic in November 2024. It has since been adopted by ChatGPT, Gemini, Microsoft Copilot, Visual Studio Code and Cursor, and in December 2025 Anthropic handed it over to the Agentic AI Foundation, part of the Linux Foundation. It works like a "universal plug": a service that publishes an MCP server becomes usable by every compatible assistant, without custom development for each one.

The most common connectors

• Build and go live: GitHub (host and version code), Vercel (publish a site and track its deployments), Supabase (database and user accounts).
• Create without coding: Lovable (generate an app from a description), Figma (read or produce mockups), Canva (visuals, posters, social media), Gamma (presentations and documents).
• Get organized: Notion, Google Drive, Gmail, Google Calendar, Slack, Airtable.

The list changes every month: each assistant shows the available connectors in its settings.

An end-to-end example

1. You describe your project to the assistant, which helps you write a good prompt.
2. Lovable or Claude Code generates the app; a Figma mockup can serve as a model.
3. The code is saved on GitHub, which keeps the history of every change.
4. Vercel puts the site online automatically after each update.
5. Canva or Gamma produce the visuals and the project presentation.

That is how IAdex itself was built: code on GitHub, hosting on Vercel.

Good practices

• Grant the minimum rights: read-only when that is enough, access limited to the relevant folders or projects.
• Approve important actions: sending an email, publishing, deleting.
• Be wary of external content: a document or page read by the assistant can contain hidden instructions (prompt injection).
• Check where your data goes: each connector passes information to the assistant; read the provider's terms, especially for personal or business data.