# Help center in your AI (MCP)

The Watbot help center has an MCP server. Connect it to your AI assistant — Claude, ChatGPT, Cursor or any other that supports the MCP protocol — and the assistant will answer questions about Watbot from the current help center and link to its pages. For example, a developer can ask how to call an API method right in the code editor, without searching in the browser.

Server address:

```
https://support.watbot.org/mcp
```

- No API key or Pro plan is needed: the server is open, just like the help center itself.
- The server knows only the help center pages. It can't see your account, bots or contacts — use the [API](https://support.watbot.org/api) to work with them.
- Transport — Streamable HTTP, no authorization.

> [!TIP]
> You can ask AI about the help center without any setup: the "Ask AI" button in the help center header opens the [AI chat](https://support.watbot.org/ai/chat) in your account with the question in the input field. Send it, and the model looks for the answer in this same help center.

## What the server can do

| Tool | What it does |
|---|---|
| `search_docs` | Searches the help center by the meaning of the query and returns matching sections with links |
| `get_page` | Returns the full text of a page by its address — with links to its sections and subpages |
| `list_pages` | Shows the table of contents of the help center or of one of its sections |

The help center is available in several languages. The assistant picks the language with the `locale` parameter; if it doesn't, the server answers in the main language of its address.

## Connecting

Services rename their menu items from time to time. If you can't find the right one, search the service's help for "custom MCP connector" or "remote MCP server".

### Claude

In Claude on the web or in the desktop app, open **Settings → Connectors** and click **Add custom connector**. Any name will do, for example "Help center"; the address is the server address above. Then turn the connector on in the tools menu under the message field.

### Claude Code

```bash
claude mcp add --transport http help https://support.watbot.org/mcp
```

### ChatGPT

ChatGPT connects custom MCP servers in developer mode: **Settings → Apps → Advanced settings → Developer mode**. Then create an app: the server address above, authentication — **No Auth**. Developer mode is not available on every ChatGPT plan.

### Cursor

Add the server to `.cursor/mcp.json` in the project folder, or to `~/.cursor/mcp.json` to have it in all projects:

```json
{
  "mcpServers": {
    "help": {
      "url": "https://support.watbot.org/mcp"
    }
  }
}
```

### VS Code

For GitHub Copilot agent mode — the `.vscode/mcp.json` file in the project folder:

```json
{
  "servers": {
    "help": {
      "type": "http",
      "url": "https://support.watbot.org/mcp"
    }
  }
}
```

### Other assistants

Any client that connects remote MCP servers over HTTP will work: enter the server address, no authorization needed. If the client can only run local servers, connect the help center through the `mcp-remote` proxy (requires Node.js):

```json
{
  "mcpServers": {
    "help": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://support.watbot.org/mcp"]
    }
  }
}
```

## Check the connection

Ask the assistant: "How do I send a mailing by days of the week? Find the answer in the help center." It should call `search_docs` and answer with a link to the help center page about mailings.

## Without connecting

- **A single page.** The "Copy" button next to the title of any page copies its text as markdown — you can paste it into any AI chat. The arrow next to the button opens ChatGPT and Claude with a question about the page right away.
- **Page text** — the same address with `.md` at the end; for the home page — `/index.md`. No formatting, menus or scripts: easier to hand to AI.
- **Table of contents for AI** — `/llms.txt` at the root of the help center: all pages with links to their text in markdown.

## Limits

- **Rate** — 60 requests per minute from one IP address. Above the limit the server responds with code `429`; the assistant can retry in a minute.
- **Search by meaning** has its own limit. Under heavy load the server searches by words only: results still come back, but are matched less precisely.