The Bedones MCP connector is a server that ChatGPT or Claude query on your behalf to read your conversations, reply to your comments and look up your catalogue. It runs on MCP (Model Context Protocol), an open standard that lets an AI call real functions inside external software, and on OAuth 2.0 for authorisation. The AI never gets your passwords: it gets access scoped to the one organisation you name.

What the MCP protocol changes for an AI

The MCP protocol plugs an AI into external software and lets it carry out actions there. Without MCP, ChatGPT and Claude stay text-only counterparts: you paste a customer conversation into the assistant window, you get a draft reply back, then you paste it by hand into WhatsApp or Facebook. The AI touches nothing, it writes.

With MCP, the assistant calls functions — the protocol calls them tools — published by the software itself. Bedones exposes an MCP server at https://api-moderator.bedones.com/mcp, and ChatGPT or Claude discover the list of available actions on their own. There is no schema for you to describe and no command syntax to memorise: discovery is part of the protocol.

That mechanism is why the Bedones MCP connector is not a proprietary plugin. Because MCP is an open standard, the same server answers ChatGPT, Claude and any other client that speaks the protocol. What you set up once still holds if you switch assistants, and the integration work does not start over with every new tool released during the year.

The tools the Bedones connector exposes to the AI

The Bedones MCP server publishes a list of tools, each covering one precise action and carrying a label that says whether it reads or changes something. The assistant therefore knows the nature of every call before it fires, and you can tell from the name of the tool used what was actually touched in your account.

  • Read: list conversations, read messages, view comments, browse products and tickets.
  • Write: send a message, reply to a comment, hide a comment, create a ticket.
  • Configuration: set your AI agent’s context, add FAQ rules, adjust moderation.

Destructive actions sit apart from that list. Deleting a comment, for instance, is reserved for organisation administrators: a member without that role cannot trigger the matching tool, even by asking the AI outright. The MCP connector mirrors the rights you already hold in Bedones, and never widens them.

Splitting read, write and configuration has an immediate practical benefit. You can start with pure read requests, along the lines of “list my unread conversations”, while you build confidence, then move on to replies that actually get sent. The full walkthrough sits in the guide to installing the connector in ChatGPT and Claude.

OAuth 2.0: you are the one who grants access

Authorising the MCP connector runs on OAuth 2.0, the same standard as the “Sign in with Google” button. You paste no password into ChatGPT or Claude. The assistant redirects you to a Bedones screen, you authenticate with us, you pick the organisation concerned, then you click “Authorise” — and access stops at that organisation.

What the AI receives at the end of that sequence is an access token, and nothing else. Your social network tokens never leave Bedones: the Meta and TikTok credentials stay server-side, and the assistant sees the result of each call but never the keys that made it possible. A leak on the assistant’s side would therefore not expose your social accounts.

Control does not end at the authorisation screen either. On every request, Bedones re-checks that the account behind the token still has the right to act on that organisation: if your role changes or you leave the team, the AI’s access follows right away. You can also revoke the connector in one click from the ChatGPT or Claude settings.

One connector for WhatsApp, Facebook, Instagram and TikTok

A single Bedones connector gives the AI access to WhatsApp, Facebook, Instagram and TikTok at the same time. You do not wire four integrations into ChatGPT or Claude and you do not track four separate authorisations: the accounts are already linked inside Bedones, and the MCP server exposes them behind one shared tool list.

That centralisation changes how you phrase requests. “Show me everything waiting for a reply” covers all four networks at once, because the conversation-reading tool does not care which platform a message came from. The same logic applies to the catalogue and to tickets, which belong to the organisation rather than to any one social network.

One connector also limits how many places your access credentials travel through. Adding each network separately to an assistant would multiply the authorisations to watch, then to revoke one by one the day you change tools. Here, a single authorisation is cut in a single move, and the links between Bedones and the platforms stay managed inside Bedones.

Turning the connector on and checking that it answers

Turning on the MCP connector takes three moves: open the connector settings in ChatGPT or Claude, add the address https://api-moderator.bedones.com/mcp as a custom connector, then authorise access on the Bedones screen by selecting your organisation. The assistant discovers the available tools on its own from there, with nothing further to configure.

The first check to run is a read request. “List my connected social accounts” or “which messages are waiting for a reply?” is enough: if the AI answers with your real data, the whole chain works, from OAuth authorisation through to the tool calls. If it answers that it has no tools available, the authorisation did not complete and needs to be started again.

The MCP connector then behaves like any other tool in your assistant, whatever market you sell into. The same calls power replying to customers in Cameroon from ChatGPT or Claude or handling messages for a shop in Côte d’Ivoire: only the content of the conversations changes, never the protocol.