Smart Search MCP

The Smart Search MCP acts as a standardized bridge between your data and AI applications. By exposing dedicated search and fetch tools, this integration empowers large language models and custom agents to natively query and read your WordPress content in real-time.

Credentials

Your unique MCP endpoint URL and MCP token is located in the WP Engine User Portal. Go to the WP Engine Smart Search product page. Identify the associated environment and click the Show Credentials button. Copy the credentials ( MCP URL and MCP Token). These will be used for authentication.

Note

A legacy SSE endpoint (/sse) is also available but new integrations should use /mcp.

Available MCP Tools (Endpoints)

The Smart Search MCP exposes the following tools to the connected client:

fetch

Fetches a specific post by its ID directly from the connected index.

Parameters:

Parameter	Required	Type	Description
id	Yes	String/Integer	The unique identifier of the post to fetch.

search

Executes a search query against the connected index. Clients should be instructed to refine search queries as much as possible to ensure relevant context windows.

Parameters:

Parameter	Required	Type	Description
query	Yes	String	The primary search query.
filter	No	String	A strict filter string supporting field targeting, boolean operators, wildcards, and ranges. See the Find API for full syntax details. Examples: field:value, field:(value1 OR value2), NOT field:value
limit	No	Integer	Limits the number of results returned in the payload.
offset	No	Integer	The offset used for pagination.

Rate Limiting

The MCP endpoints utilize a token bucket algorithm to handle traffic and protect against brute-force attacks.

1. Request Rate Limiter (General API Traffic)

This limiter governs successful requests to the MCP tools.

Custom Code / LLM integrations:

• Rate: 5 Requests Per Second (RPS)
• Burst Capacity: Up to 30 requests
• Scope: IP-based
• Behavior: The bucket starts with 30 tokens. Each request consumes one token. Tokens refill at a steady rate of 5 per second. If the bucket hits 0, requests are rejected until the bucket refills at the steady RPS rate.

ChatKit integrations:

• Rate: 10 Requests Per Minute
• Scope: OpenAI session-level

Unauthenticated access:

• Rate: 5 RPS
• Burst Capacity: 10
• Scope: Shared globally to all usages

Note

Unauthenticated access is not allowed for ChatKit integration.

2. Auth Failure Limiter

This limiter specifically protects against brute-force authentication attempts.

• Rate: 1 allowed failure per 30 seconds
• Burst Capacity: 2 failed attempts
• Scope: IP-based

Connection Configuration

Use your MCP URL and MCP token to configure the connection for your client. For platforms that do not support authentication headers, limited unauthenticated access is available.

ChatKit & Agent Builder

See the ChatKit configuration guide for setup details.

Claude Code (Terminal)

Add the server directly via the command line using the Claude MCP add command with the http transport flag. Replace {your-mcp-url} with your /mcp URL and {your-mcp-token} with your MCP token.

claude mcp add --transport http search_mcp "{your-mcp-url}" --header "Authorization: Bearer {your-mcp-token}"

Claude (Web Version) & ChatGPT Desktop (Developer Mode)

Configure your remote servers within their respective environments by using your /mcp URL. For platform-specific setup, see the Claude remote MCP guide and ChatGPT Developer Mode guide.

Caution

Do not provide an authentication header, as this authentication configuration is not supported by those platforms. Please refer to Anthropic’s and OpenAI’s official platform guides for modifying your configuration.

Custom Code (e.g., Headless WP)

Include your bearer token in the Authorization header of each request.

Authorization: Bearer {your-mcp-token}

For implementation examples, see Smart Search MCP in Headless WP and Using MCP Servers in Concert.