---
title: "MCP Server - Serpwise Documentation"
description: "Expose a Model Context Protocol endpoint on your domain so AI agents can query site content and products through structured tools."
url: "https://serpwise.ai/docs/mcp-server/"
source: "https://serpwise.ai/docs/mcp-server/"
---
[Back to Serpwise](/)

Serpwise Documentation

Connect · Configure · Execute

Search documentation

`/`

No documentation found.

↑↓ Navigate Enter to open

Browse documentation

**MCP Server** exposes a [Model Context Protocol](https://modelcontextprotocol.io/) endpoint directly on your domain at `/mcp`. AI agents connect to this endpoint and use structured tools to search your content, browse your product catalog, and retrieve page details - all without scraping HTML.

When enabled, the edge proxy intercepts `/mcp` requests, authenticates the caller, and processes them securely. Your origin server never sees these requests.

* * *

## How It Works

```
AI Agent (Claude Desktop, ChatGPT, etc.)
  │
  ▼
customer-site.com/mcp  ──►  Serpwise Edge Proxy  ──►  Semantic Search & Content Engine
                              │
                              ▼
                         /.well-known/mcp.json  (auto-discovery)
                         /llms.txt              (AI agent discovery)
```

1. An AI agent discovers the MCP endpoint via `/.well-known/mcp.json` or `/llms.txt`
2. The agent sends an MCP initialization request to `customer-site.com/mcp`
3. The edge proxy authenticates the request via API key
4. The proxy processes the tool call (semantic search, page retrieval, etc.)
5. Results are returned to the agent as structured JSON

* * *

## Available Tools

| Tool                  | Description                                    | Parameters                                      |
|-----------------------|------------------------------------------------|-------------------------------------------------|
| `search_content`      | Semantic search across all crawled pages       | `query` (required), `domain_id`, `limit` (1-50) |
| `get_page`            | Get full markdown content of a specific page   | `url_path` (required), `domain_id`              |
| `search_products`     | Semantic search across the product catalog     | `query` (required), `domain_id`, `limit` (1-50) |
| `get_product_details` | Full enriched product data (30+ AI attributes) | `product_id` (required)                         |
| `list_categories`     | Browse all product categories                  | `domain_id`                                     |

All queries are automatically scoped to your organization and domain - no cross-account data access is possible.

* * *

## Enabling MCP Server

1. Navigate to your domain in the dashboard
2. Click the **AI** tab
3. Select the **MCP Server** sub-tab
4. Toggle **Enable MCP Server**
5. Click **Save MCP Settings**

Changes take effect immediately. The `/mcp` endpoint, `/.well-known/mcp.json`, and `/llms.txt` become active as soon as you save.

* * *

## Discovery Endpoints

When MCP is enabled, two discovery endpoints are automatically served:

### `/.well-known/mcp.json`

JSON manifest describing the MCP server, its endpoint URL, authentication requirements, and available tools. AI agents use this for automatic discovery.

### `/llms.txt`

Plain text file following the emerging `llms.txt` standard. Describes the site and points AI agents to the MCP endpoint for structured access.

* * *

## Authentication

MCP requests require a **Bearer token** - an API key with the `products:read` scope.

1. Go to **Organization Settings** → **API Keys**
2. Create a new key with the `products:read` scope
3. Include it in MCP requests as: `Authorization: Bearer sw_prod_...`

API keys can optionally be restricted to specific domains.

* * *

## Connecting from Claude Desktop

Add this to your Claude Desktop MCP configuration (`claude_desktop_config.json`):

```
{
	"mcpServers": {
		"my-site": {
			"url": "https://example.com/mcp",
			"headers": {
				"Authorization": "Bearer sw_prod_your_api_key_here"
			}
		}
	}
}
```

Restart Claude Desktop. The five tools will appear in the tool picker.

* * *

## Connecting via curl

Initialize a session:

```
curl -X POST https://example.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json" \
  -H "Authorization: Bearer sw_prod_your_api_key_here" \
  -d '{
    "jsonrpc": "2.0",
    "method": "initialize",
    "params": {
      "protocolVersion": "2025-03-26",
      "capabilities": {},
      "clientInfo": { "name": "my-agent", "version": "1.0" }
    },
    "id": 1
  }'
```

Call a tool:

```
curl -X POST https://example.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json" \
  -H "Authorization: Bearer sw_prod_your_api_key_here" \
  -d '{
    "jsonrpc": "2.0",
    "method": "tools/call",
    "params": {
      "name": "search_content",
      "arguments": { "query": "return policy", "limit": 5 }
    },
    "id": 2
  }'
```

* * *

## Cloudflare Configuration

If your domain uses Cloudflare, AI agents will be blocked by Cloudflare’s bot challenge (they can’t execute JavaScript). You need to create a WAF rule to skip the challenge for MCP requests:

1. Go to **Cloudflare Dashboard** > **Security** > **WAF** > **Custom Rules**
2. Create a new rule:
   
   - **Name**: Allow Serpwise MCP & AI Discovery
   - **If**: URI Path starts with `/mcp` **OR** URI Path equals `/.well-known/mcp.json` **OR** URI Path equals `/llms.txt`
   - **Then**: Skip (All remaining custom rules, or set to Allow)
3. Save and deploy

This ensures AI agents can reach the MCP endpoint and auto-discover it via the standard discovery files.

* * *

## Rate Limits

MCP requests are subject to the API key’s rate limits (configured per key):

- **Per minute**: Default 60 requests
- **Per day**: Default 10,000 requests

Rate limit headers (`X-RateLimit-Limit`, `X-RateLimit-Remaining`, `X-RateLimit-Reset`) are included in responses.