Run your AI front desk from ChatGPT or Claude via MCP

What the front-desk MCP server gives your AI

MCP (Model Context Protocol) is an open standard that lets an AI assistant call tools of an external service. The BotB2B front-desk server exposes the same things you see in Front Desk on the site, grouped into five facades. Your AI sees them as tools and picks the right one for the task.

The server address is the same for every client: https://api.botb2b.ai/api/mcp/front-desk. Access is granted through OAuth: the AI never sees your bot tokens or passwords, and you can revoke it in one click under AI Staff → MCP for any AI agent in the app.

Step 1. Connect the server to your AI

1. 1

   Prepare the account

   Sign up at app.botb2b.ai and connect at least one channel, for example a Telegram bot or the website widget. The AI will assign bots to these channels.

   Click to enlarge

2. 2

   Add the connector in Claude

   Open claude.ai → Settings → Connectors → Add custom connector, paste https://api.botb2b.ai/api/mcp/front-desk and press Add. Then press Connect: a BotB2B window opens, you sign in and choose which accounts the AI may see.

   Other clients: ChatGPT → Settings → Connectors → Add (developer mode required); Cursor → ~/.cursor/mcp.json; Claude Code → claude mcp add --transport http botb2b-front-desk https://api.botb2b.ai/api/mcp/front-desk. Details in How to connect MCP.

   Click to enlarge

3. 3

   Check the tools

   In a new chat ask: "What BotB2B tools do you have?" The AI lists the facades: bots, chats, leads, knowledge base, integrations. In ChatGPT you also need to switch the connector on in the chat with the + button.

   Click to enlarge

Step 2. Let the AI build the bot

The strongest way to start is to let the AI learn from what already happened. Paste prompts like these into the chat; the AI reads the conversations through the Chats facade and writes the bot through Bots and Knowledge base.

• "Study my conversations from the last 3 months and build an assistant profile: tone of voice, frequent questions, the answers we give."
• "Create a front-desk bot for my store based on this profile. Put facts about prices, delivery and payment into a knowledge base and link it to the bot."
• "Set the bot to ask for help when the customer insults, asks for a manager or wants a discount over 20 %."
• "Assign the bot to my Telegram account and start it."

No history yet? Describe the business yourself: who you are, what you sell, prices, delivery, what to ask the customer. The AI turns it into the four scenario blocks explained in Bot settings: every field.

Step 3. Test before customers do

The Chats facade can create a test conversation and send messages as a simulated customer, so the AI can play several roles and show you the transcripts.

• "Test the bot: write to it as an unhappy customer, as a wholesaler and as someone just asking. Show me the dialogs."
• "Try to get a discount and to talk the bot off topic. Where did it break?"
• "Fix the prompt so that the bot always asks for a phone number before quoting delivery."

Step 4. Route leads and calls for help

Through Leads and Integrations the AI sets where captured contacts go and who picks up conversations the bot could not finish.

• "Set it up so that leads qualified as hot land in my CRM pipeline Sales, stage New."
• "Tag every lead from the last week that mentions delivery outside the city with the tag Delivery."
• "Show conversations where the bot asked for help and assign them to Anna."

Telegram and email notifications for leads are configured on the bot page, see Notifications and CRM; the duty schedule for the team lives in Team schedule.

Step 5. Review and retrain weekly

• "Review the bot's dialogs for the week: where did it answer poorly and what should I fix in the knowledge base?"
• "Compare the conversion of my two bots over the last month and suggest what to change in the weaker one."
• "Add these five questions and answers to the knowledge base and update the bot."

Want it done without you in the loop? The AI manager uses the same tools on a schedule and reports in Telegram. MCP is you at the wheel; the AI manager is a task you hand over.

What the AI cannot do and how to stay safe

• The AI sees only the accounts you ticked during the OAuth consent; keys and tokens stay on the platform
• Destructive actions such as deleting a bot or a knowledge base are ordinary tools: ask the AI to confirm before deleting, or keep such prompts explicit
• The AI cannot connect new channels for you: adding a Telegram bot or a personal number needs your token or QR code on the site
• Access is revoked under AI Staff → MCP for any AI agent in the app or in the client's connector settings; tokens are not affected

Click to enlarge