Claude MCP: Run Your Website Agent From Claude

Claude MCP connects Claude to outside software through a custom connector, and the connector most guides skip is the one a business can use today: its own customers. Paste https://mcp.ultimo-bots.com/mcp under Connectors, sign in, and Claude can read every question your website AI agent was asked this week. After the setup comes the part developer guides skip: what to ask on a Monday, and how one message to Mo turns the answer into a booking that happens while your shop is closed.

What is MCP for Claude?

MCP (Model Context Protocol) is the open standard that lets Claude use outside software: a server lists what an AI may do there, and Claude calls those tools when a chat needs them. Anthropic open-sourced it in November 2024. So yes, Claude can use an MCP server, and you do not have to run one. The vendor hosts it, and you only add its address.

What are Claude connectors?

A Claude connector is a remote MCP server added to your Claude account. You add it once on the web or in Claude Desktop, and Anthropic's help article says it then works across claude.ai, Claude Desktop, Cowork and the mobile apps (read November 8, 2026). In each chat you turn it on with the + button under Connectors, next to others such as Gmail or Google Calendar. One of them can be the AI agent on your website.

Good morning

How can I help you today?

Add files or photosConnectors

ConnectorsUltimo BotsGmailGoogle CalendarManage connectors

Ultimo Bots is the AI agent platform for people who would rather describe a job than configure one: you tell Mo, the builder in the AI Studio, what your agent should do, and Mo builds it on your apps, public APIs and remote MCP servers. The agent works on your website, WhatsApp, Instagram and more. Ultimo Bots also runs its own MCP server, and that server is the connector this guide adds.

When should you use an MCP server?

Use an MCP server when Claude needs live data from another system, or has to change something there. Without one, you export, paste and copy the result back by hand, and few people keep that up. With one, "what did customers ask this week?" is a single sentence that reads the real chats. For a one-off question about a document you already have, attach the document and skip the server.

How do I add an MCP server to Claude?

You add a remote MCP server once under Customize and Connectors, and Claude Code takes one command. Five minutes is enough. There is no config file to edit and no key to copy.

In claude.ai, Claude Desktop and Cowork

1. 1

   Open Connectors

   Go to Customize, then Connectors. Click + Add and choose Add custom connector. On Team and Enterprise an owner does this first under Organization settings, and each member then clicks Connect.

2. 2

   Enter the server

   Name it Ultimo Bots, enter the URL and click Continue. Review the detected settings and click Continue again.

   https://mcp.ultimo-bots.com/mcp

   Copy
3. 3

   Choose the sign-in

   Under Authentication choose Sign in now. Under OAuth client choose Register automatically. Leave the request headers empty and click Add.

4. 4

   Approve on the consent page

   Sign in to Ultimo Bots and choose a workspace. Every permission starts ticked, so untick what Claude should not have, then click Allow connection.

In Claude Code

Claude Code MCP servers have their own command. Run the one below in your terminal, type /mcp, choose ultimo-bots and select Authenticate; the same consent page opens in your browser. Add --scope user to have it in all your projects.

claude mcp add --transport http ultimo-bots https://mcp.ultimo-bots.com/mcp

Team setups and every other option are in our Claude guide in the help center.

What can Claude do once it is connected?

Claude can read your conversations and leads, show your insights, add knowledge and change the widget, live chat and notification settings, all in plain language. The server has more than 100 tools, and Claude can only use the ones your approval covers. The real gain is what Claude does with them: it can weigh a whole week of chats at once and then write for you.

Take Alder Street Bike Works, a fictional repair shop with one mechanic and one agent on its website. On Monday morning, between two repairs, Dana, the owner, opens Claude on her phone, turns on the connector and types:

Dana answers in the same chat: "Add a Q&A: we service e-bikes but do not replace batteries. Then show me the new leads and draft a short reply to each." Claude adds the Q&A with add_knowledge_source, reads the leads with list_leads and get_lead, and notices that Leo needs his bike for a race on the 14th, so his reply goes first. Dana sends the replies herself. None of this uses AI credits, because the thinking runs on her Claude plan. The e-bike line was a quick fix. The tune-up question is different: five people wanted to book and got a phone number, and more knowledge will not change that. Her agent needs to do the booking.

Turn what Claude finds into a skill with Mo

Claude shows you what your agent is missing, and Mo builds it. Skills take shape in the AI Studio: you describe the job in plain words, typed or dictated, and Mo builds and tests it. Claude can even write the brief. Dana asks it for "the message I should send my agent builder so those five people could book", checks the result and pastes it into the AI Studio.

Cal.com is among the ready-made skills, tested templates for apps such as Cal.com, Stripe and Wix that Mo adjusts for you. Mo shows a Connect button for Dana's Cal.com account, sets up the booking and tests it with questions that only read. If Dana tries a booking herself in the Test tab, it is a real one, so she cancels it afterwards.

Shop openShop closed

MonClaude spots what's missing

TueMo builds the skill

Wed

Thu

Fri

Sat

SunSlot picked at 11:40 pm

The next Sunday at 11:40 pm, with the shop dark and Dana asleep, a visitor asks whether anyone can look at a slipping chain this week. Under the widget's AI notice, the agent offers Thursday at 9:00 am, reads the details back and, after a yes, answers: "Thursday at 9:00 am for Jo: the confirmation from Cal.com is on its way to your inbox." On Monday Dana asks Claude the same question as last week, and the list is shorter.

What Dana can tell Mo next

• "Let people move or cancel their booking." Another ready-made skill. The visitor confirms their email with a 6-digit code first, and the agent changes a booking only after a yes.
• "Sell our gift vouchers." The ready-made Stripe skill shows her products and prices and hands over a secure payment link, so card details never enter the chat.
• "Check our parts supplier's API before you promise a chain is in stock." For a system with no ready-made skill, Dana sends Mo the API docs or an MCP server address, and the key goes into a secure field.

Each of those counts as one skill, so booking, changes and vouchers fit into Smart: $29 a month ($19 billed yearly) with 500 AI credits and 3 skills. Boost is $59 ($39) with 2,000 AI credits and 10 skills, Ultimo $249 ($199) with 10,000 and 30 skills. One balance per agent pays for its answers and for Mo, and teammates are free. See pricing and what else Mo can build in the AI Studio.

Is Claude MCP free?

The protocol is free, and Anthropic offers custom connectors on every Claude plan, with Free limited to one (read November 8, 2026). Connecting Ultimo Bots adds no fee on our side and works on any Ultimo Bots plan. Claude's usage limits apply to the chat itself, and our AI credits pay only for your agent's answers to visitors and for Mo.

What stays in your control?

You decide on one consent page what Claude may touch, and you can end it at any time. You choose one workspace, the permissions, and Only selected agents or All agents in this workspace, including future agents. Permissions marked ⚠ can cause costs, send messages in your name or delete data. They all start ticked, so untick the ones you do not want before you click Allow connection. Your own role still applies, so Claude can never do more in your workspace than you could yourself.

• Claude asks as well: Anthropic advises reading Claude's tool approval requests before you click Allow always, and that is good advice for anything that writes.
• How long it lasts: the connection stays active while you use it and ends after 30 days without use, after a year at the latest.
• Where to end it: on the MCP connections page in the portal, next to Agents and Workspaces, Revoke access cuts Claude off at once.
• Visitors' words: conversations and leads hold what strangers typed. Read what Claude proposes before you approve a change that came out of them.

Ten minutes, one question

Give Claude ten minutes and a question that changes nothing. Add https://mcp.ultimo-bots.com/mcp, click Clear all on the consent page, tick only the read permissions, and ask: "What did visitors ask this week that my agent could not answer?" You end up with a short list in your customers' own words. A line you can fix with a sentence goes into your knowledge base. A line that starts with "can I book" goes to Mo. Still without an agent? Start one in the portal and come back after its first week of questions. We wrote separate guides for Codex, Cursor, ChatGPT and Symphony by Wix.

Questions and answers