> ## Documentation Index
> Fetch the complete documentation index at: https://sitespeak.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Expose Your Knowledge Base via MCP

> Let AI agents like Claude, Cursor, and Windsurf query your chatbot's knowledge base directly using the Model Context Protocol. Includes automatic WebMCP browser discovery.

Every SiteSpeakAI chatbot can act as an MCP (Model Context Protocol) server, giving AI agents direct access to your knowledge base. Instead of scraping your website or chat widget, agents connect to a structured endpoint and get accurate, sourced answers from your trained content.

SiteSpeakAI supports two ways for AI agents to access your knowledge base:

1. **MCP Server endpoint** - AI tools like Claude Desktop, Cursor, and Windsurf connect directly to your chatbot's MCP endpoint. Requires manual configuration in each tool.
2. **WebMCP (browser discovery)** - AI agents running in the browser automatically discover your knowledge base through the chatbot widget. No configuration needed.

<Note>
  MCP Server and WebMCP are available on **Pro**, **Pro Plus**, and **Business** plans. Each plan includes a monthly query allowance (Pro: 1,000 / Pro Plus: 2,500 / Business: 10,000).
</Note>

## Why expose your knowledge base via MCP?

AI agents are becoming the primary way people find and interact with information. When a developer asks Cursor about your API, or a user asks Claude about your product, those agents need a reliable source of truth. Your MCP server provides that.

* **Reach AI agents directly.** Your knowledge base becomes a tool that any MCP-compatible agent can call, not just visitors to your website.
* **Accurate, sourced answers.** Agents get results from your trained content with source URLs - no hallucinations from outdated web scrapes.
* **Full RAG pipeline.** Every query runs through semantic search, query expansion, and reranking for high-quality results.
* **Control access.** Choose between authenticated access with API tokens or public access for broader reach.
* **Track usage.** Monitor MCP query usage alongside your other metrics in the dashboard.

## How it works

When an AI agent connects to your MCP server endpoint, it gets access to one tool:

**`query_knowledge_base`** - Searches your chatbot's trained content using the full RAG pipeline (vector search, query expansion, and Cohere reranking). Returns relevant content chunks with source URLs.

The agent sends a natural language query, and your MCP server returns the most relevant content from your training sources - the same content your chatbot uses to answer questions.

## Enable MCP Server

<Steps>
  <Step title="Go to Settings">
    In your chatbot dashboard, select your chatbot and go to **Settings**.
  </Step>

  <Step title="Open Advanced settings">
    Click the **Advanced** tab.
  </Step>

  <Step title="Find MCP Server">
    Scroll down to the **MCP Server (Public Access)** section.
  </Step>

  <Step title="Enable public access (optional)">
    Toggle **Public MCP Access** to allow AI agents to query your knowledge base without authentication.

    When disabled, agents need a valid API token to connect (see [Authentication](#authentication) below).
  </Step>

  <Step title="Copy your endpoint URL">
    Click the copy button next to your **MCP Endpoint** URL. This is the URL AI agents will use to connect.

    Your endpoint URL follows this format:

    ```
    https://chatbot.sitespeak.ai/api/{chatbotId}/server/mcp
    ```
  </Step>

  <Step title="Save settings">
    Click **Save** to apply your changes.
  </Step>
</Steps>

## WebMCP - automatic browser discovery

WebMCP lets AI agents running in the browser automatically discover and query your knowledge base, with zero configuration from the end user. When someone visits your website with a WebMCP-compatible browser or AI assistant, the chatbot widget registers itself as a tool the agent can use.

### How WebMCP works

When your chatbot widget loads on a page, it checks for `navigator.modelContext` support in the browser. If available, the widget automatically registers a tool called `query_{your_chatbot_name}` that the browser's AI agent can call.

When an AI agent calls this tool:

1. The query is sent to your chatbot's search pipeline (vector search, query expansion, and reranking)
2. The raw search results are returned directly to the agent - no LLM call is made on the server
3. The browser's AI agent uses its own language model to synthesize a final answer from the results

This is different from the standard MCP server endpoint, where the full RAG pipeline (including the LLM response) runs server-side. With WebMCP, the browser's AI agent handles the synthesis step, making responses faster and more contextual to the user's conversation.

### Requirements

WebMCP requires two things:

* **Public MCP Access enabled** - The same toggle used for the MCP server endpoint also enables WebMCP.
* **Chatbot widget installed** - The widget must be embedded on your website. WebMCP is registered automatically as part of the widget script.

No additional setup is needed. Once both conditions are met, any visitor using a WebMCP-compatible browser will have automatic access to your knowledge base through their AI agent.

### Enable WebMCP

<Steps>
  <Step title="Go to Settings">
    In your chatbot dashboard, select your chatbot and go to **Settings**.
  </Step>

  <Step title="Open Advanced settings">
    Click the **Advanced** tab.
  </Step>

  <Step title="Enable Public MCP Access">
    Toggle **Public MCP Access** on. This enables both the MCP server endpoint and WebMCP browser discovery.
  </Step>

  <Step title="Save settings">
    Click **Save** to apply your changes.
  </Step>
</Steps>

That's it. As long as your chatbot widget is installed on your website, WebMCP will work automatically for any visitor using a compatible browser.

<Tip>
  WebMCP queries use the same monthly MCP query credits as the MCP server endpoint. Monitor your usage on the [Usage page](https://sitespeak.ai/user/usage).
</Tip>

## Authentication

Your MCP server supports two access modes:

### Public access (no authentication)

When **Public MCP Access** is enabled, any AI agent can query your knowledge base without a token. This is the simplest option and works well when you want maximum reach - for example, letting anyone's Claude or Cursor instance access your documentation.

### Authenticated access (API token)

When Public MCP Access is disabled, agents must include a valid API token in the `Authorization` header:

```
Authorization: Bearer YOUR_API_TOKEN
```

Generate an API token from your [Account page](https://sitespeak.ai/user/api-tokens). This is the same token used for the [SiteSpeakAI REST API](/api-reference/introduction).

<Tip>
  Use authenticated access when your knowledge base contains internal or sensitive information. Use public access when you want to maximize the reach of your documentation or support content.
</Tip>

## Connect AI tools to your MCP server

### Claude Desktop

Add your MCP server to Claude Desktop's configuration file:

<Tabs>
  <Tab title="macOS">
    Edit `~/Library/Application Support/Claude/claude_desktop_config.json`:

    ```json theme={null}
    {
      "mcpServers": {
        "my-knowledge-base": {
          "type": "url",
          "url": "https://chatbot.sitespeak.ai/api/{chatbotId}/server/mcp",
          "headers": {
            "Authorization": "Bearer YOUR_API_TOKEN"
          }
        }
      }
    }
    ```
  </Tab>

  <Tab title="Windows">
    Edit `%APPDATA%\Claude\claude_desktop_config.json`:

    ```json theme={null}
    {
      "mcpServers": {
        "my-knowledge-base": {
          "type": "url",
          "url": "https://chatbot.sitespeak.ai/api/{chatbotId}/server/mcp",
          "headers": {
            "Authorization": "Bearer YOUR_API_TOKEN"
          }
        }
      }
    }
    ```
  </Tab>
</Tabs>

Restart Claude Desktop. You should see your knowledge base listed as an available tool.

<Note>
  If you enabled **Public MCP Access**, you can omit the `headers` field entirely.
</Note>

### Cursor

Add your MCP server in Cursor's settings:

1. Open **Cursor Settings** (Cmd+Shift+J on macOS, Ctrl+Shift+J on Windows)
2. Go to **MCP**
3. Click **+ Add new MCP server**
4. Enter the following:
   * **Name:** Your knowledge base name
   * **Type:** `sse`
   * **URL:** `https://chatbot.sitespeak.ai/api/{chatbotId}/server/mcp`

If using authenticated access, add the Authorization header in the server configuration.

### Windsurf

Add your MCP server in Windsurf's MCP configuration:

1. Open the **Command Palette** (Cmd+Shift+P on macOS, Ctrl+Shift+P on Windows)
2. Search for **MCP: Configure MCP Servers**
3. Add your server:

```json theme={null}
{
  "mcpServers": {
    "my-knowledge-base": {
      "serverUrl": "https://chatbot.sitespeak.ai/api/{chatbotId}/server/mcp"
    }
  }
}
```

### Custom MCP clients

Any MCP-compatible client can connect using the Streamable HTTP transport:

* **Endpoint:** `https://chatbot.sitespeak.ai/api/{chatbotId}/server/mcp`
* **Transport:** Streamable HTTP (POST for requests, GET for SSE streaming)
* **Authentication:** Bearer token in the `Authorization` header (when public access is disabled)

## Rate limits and query credits

MCP server requests are rate-limited to **30 requests per minute** per IP address. This is separate from your chat widget rate limits.

Each query to `query_knowledge_base` uses one query credit from your monthly allowance:

| Plan     | Monthly MCP queries |
| -------- | ------------------- |
| Pro      | 1,000               |
| Pro Plus | 2,500               |
| Business | 10,000              |

Track your usage on the [Usage page](https://sitespeak.ai/user/usage) in your account settings.

When your monthly limit is reached, the server returns a `429` status code. Credits reset at the start of each billing cycle.

## Use cases

### Documentation search for developers

Expose your product documentation so developers using Cursor, Windsurf, or Claude can look up your APIs, configuration options, and guides without leaving their editor.

### Internal knowledge base

Connect your company's trained chatbot to your team's AI tools. Sales teams can query product specs from Claude, support teams can look up troubleshooting guides, and engineers can search internal docs - all through the tools they already use.

### Customer-facing AI access

Enable public access so your customers' AI agents can query your knowledge base directly. When someone asks Claude or ChatGPT about your product, their agent can pull accurate answers from your trained content instead of relying on potentially outdated web data.

## Troubleshooting

### Agent can't connect

* Verify the endpoint URL is correct and includes your chatbot ID
* Check that your plan supports MCP Server (Pro or higher)
* If using authenticated access, confirm your API token is valid

### Getting 401 Unauthorized

* Ensure the `Authorization` header uses the format `Bearer YOUR_API_TOKEN`
* Generate a new token from your [Account page](https://sitespeak.ai/user/api-tokens) if the current one may have been revoked
* If you want to skip authentication, enable **Public MCP Access** in your chatbot settings

### Getting 429 Too Many Requests

* You've hit either the per-minute rate limit (30/min) or your monthly query credit limit
* Check your usage on the [Usage page](https://sitespeak.ai/user/usage)
* Consider upgrading your plan for a higher monthly allowance

### Results seem incomplete or irrelevant

* Make sure your chatbot has been trained with relevant content
* Check that your training sources are up to date (use auto-sync if available on your plan)
* Try rephrasing the query - the search uses semantic matching, so natural language questions work best

***

<Card title="Ready to automate your customer service with AI?" icon="bot" href="https://sitespeak.ai/register?utm_source=docs&utm_medium=cta&utm_campaign=primary-cta" arrow="true" cta="Create Your AI Agent">
  Join over 1000+ businesses, websites and startups automating their customer service and other tasks with a custom trained AI agent.
</Card>