Building in Claude, ChatGPT and other AI apps#

Chatbots, mini apps, mini landing pages and everything else you build in the AI chat can also be built and edited in your own AI client — Claude, ChatGPT, Claude Code, Codex, Cursor and others. The tools are the same as for building in the AI chat, and they connect over the MCP protocol (Model Context Protocol).

What you need#

• Pro or trial Pro for the project owner. You can connect without it, but the AI will not start building: to every request it will reply that Pro is needed and give a link to get it. A team member does not need their own subscription — the project owner's subscription is what counts. More on the The Pro plan page.
• Access to the project's AI chat. The owner always has it; a team member needs the "Full chat access" role permission in the "AI agents / Chat" group. Permissions are described on the Team and access page.
• An AI client that supports MCP over HTTP. The model runs in it and on its plan.

Your "AI Credits" are not used: the model and tokens belong to your AI service, and the building tools come with Pro.

Where to get the URL#

Open the "AI models" section in the project menu and go to the "MCP connection" page. It has the MCP server URL with a "Copy" button and instructions for each client on a separate tab: Claude, ChatGPT, Claude Code, Codex, Cursor, VS Code, LM Studio and "Other".

There is one URL for the whole account. The project is chosen later, in the conversation with the AI.

How to connect#

Most clients sign in to the account by themselves, over OAuth: after you add the server, a sign-in window opens where you sign in to Watbot and allow access. You do not need an API key for this.

Claude#

1. Open Claude on the web or in the app → "Settings" → "Connectors".
2. Click "Add custom connector", enter a name and the MCP server URL.
3. Click "Connect", sign in to your account and allow access.
4. In a new chat, turn the connector on in the tools menu.

ChatGPT#

1. In ChatGPT, open "Settings" → "Apps & Connectors" → "Advanced settings" and turn on developer mode.
2. In "Apps & Connectors", click "Create", enter a name, the MCP server URL and OAuth as the authentication.
3. Sign in to your account, allow access and pick the connector in a new chat.

Claude Code and Codex#

The server is added with a terminal command — the ready-made command with your URL is on the client's tab. In Claude Code, after running it, type /mcp, pick the server and sign in to your account. In Codex, a second command, codex mcp login, opens the sign-in; the server appears in Codex CLI, in the IDE extension and in the Codex app.

Cursor and VS Code#

The server is added to a settings file: in Cursor — ~/.cursor/mcp.json, in VS Code — the project's .vscode/mcp.json. The ready-made file text is on the client's tab. Then sign in to your account: in Cursor — with the "Login" button next to the server in the MCP settings, in VS Code — with the "Start" button above the server in the file.

LM Studio and other clients#

Clients that cannot sign in over OAuth connect with an API key — it is passed in the Authorization: Bearer <key> request header. On the "LM Studio" and "Other" tabs, the example has the <API token> placeholder, and the "Copy" button puts your key in its place. If you do not have a key yet, one is created; a key that has already been issued does not change.

Warning

The API key gives access to the whole account. Do not show it to others and do not publish settings files that contain it.

How to work#

Tell the AI what to build: "build a bot for booking at a beauty salon", "add a request form to the landing page". From there the AI acts the same way as in the AI chat:

1. Picks the project. If you have one project, it is picked automatically. If you have several, the AI shows the list and asks which one to work in. The list has your projects and the ones where you are on the team.
2. Turns on the right builder — bot, mini app, landing page and so on — and gets its rules.
3. Builds and edits. Changes are saved right away. There are no "Apply" cards here: permission for each call is asked by the AI client itself, if that is turned on in its settings.
4. Gives a link. After building, the AI shows a link to the editor in your account — the bot scheme, the landing page, the mini app version — where you can open and check what was built.

Each conversation in the client is a separate session: the project and bot chosen in one conversation do not affect another. Such conversations do not appear in the chat list of the "AI chat" section.

What is done only in your account#

Over MCP the AI does not perform actions that always wait for your confirmation in the AI chat:

• publishing mini landing pages, mini app versions and courses;
• deleting documents from a knowledge base;

You do these yourself in your account — the AI will tell you where. Creating projects, connecting messengers and launching mailings over MCP is not possible either.

If something does not work#

• The AI replies that Pro is needed. The project owner does not have Pro or trial Pro. The AI gives the owner a link to get it; a team member should ask the owner.
• The AI says there is no access to the project's AI chat. Your role in this project does not have the "Full chat access" permission — ask the project owner to add it.
• The project is not in the list. The list has only your projects and the projects where you are on the team. Projects of a blocked owner are not shown.
• The client cannot sign in to the account. Check that the server URL was copied in full. If the client does not support OAuth, connect it with an API key, as on the "Other" tab.
• A client connected with an API key stopped working. The key was reissued in the profile — the old one stops working right away. Copy the example on the "MCP connection" page again.