DocumentationAEMAEM as a Cloud Service Tutorials

Search and analyze AEM content with natural language

Last update: October 26, 2026
  • Applies to:
  • Experience Manager as a Cloud Service

CREATED FOR:

  • Beginner
  • Leader
  • User
  • Developer

Use the Content AI MCP Server, a companion to the AEM MCP Server, from Adobe CX Enterprise Coworker to search and analyze Content AI content sources in natural language, no low-level API code or UI navigation.

In this tutorial you discover available content sources, run keyword, semantic, and hybrid searches, and use generative search to get a synthesized answer, all from Coworker against a Content AI content source.

Tools and access

The Content AI MCP Server provides six tools:

Tool
Purpose
list_content_sources
Discover available content sources in your AEM environment.
get_content_source_config
Inspect the raw content source configuration and metadata.
fulltext_search
Keyword-based search with fuzzy matching and field selection.
semantic_search
Vector-based similarity search for conceptual queries.
hybrid_search
Combined keyword + semantic search for the best general-purpose recall.
gen_search
Generate a synthesized answer from matching content, instead of returning raw documents.

The Content AI MCP Server is public-source-only: the X-Api-Key header grants source access. An API key is required; a request without it is rejected.

  • Public (anonymous): Search public, read-only content sources using the X-Api-Key header. Use this for openly available, non-sensitive content. See Searching public content sources.
  • Authenticated: Search entitled or access-controlled content sources using the x-content-ai-mcp-api-key header together with your signed-in Adobe identity (passed through by CX Coworker). Results respect your permissions.

Add the Content AI MCP Server

Connect Adobe CX Enterprise Coworker to the Content AI MCP Server and run the scenario below.

Let’s set up the Content AI MCP Server in Coworker with these steps.

  1. Sign in to Coworker.

    Adobe CX Coworker Sign In {modal="regular"}

  2. From the side rail, click MCP Servers.

    MCP Servers Side Rail {modal="regular"}

  3. Click Add MCP Server.

    Add MCP Server {modal="regular"}

  4. In the Add MCP Server dialog, enter the following details:

    table 0-row-2 1-row-2 2-row-2 3-row-2 4-row-2
    Field Value
    Server name content-ai-mcp
    MCP URL https://mcp.adobeaemcloud.com/adobe/experimental/aemagents-expires-20260331/mcp/content-ai
    Connection type HTTP
    Authentication type Passthrough

    Add MCP Server Dialog {modal="regular"}

    note
    NOTE
    Authentication type is set to Passthrough so Coworker forwards your signed-in Adobe identity to the MCP Server. You do not need a separate sign-in or OAuth step.

  5. Click Add headers and add the headers for the access mode you need:

    For public content sources (anonymous):

    table 0-row-2 1-row-2 2-row-2
    Header Value
    X-Api-Key Your public API key. Present this header when you only want to search public content sources.
    x-content-ai-mcp-routing (Optional) Environment/routing info, for example tier=publish,bucket=p12345-e67890,cluster=ethos21-prod-va7,namespace=ns-team-example.

    For non-public (entitled) content sources:

    table 0-row-2 1-row-2 2-row-2
    Header Value
    X-Api-Key Your Content AI API key. Required — searches are rejected without it.
    x-content-ai-mcp-routing (Optional) Environment/routing info, for example tier=publish,bucket=p12345-e67890,cluster=ethos21-prod-va7,namespace=ns-team-example.
    note important
    IMPORTANT
    When the X-Api-Key header is present, the server treats the request as anonymous and only searches public content sources. Any signed-in identity is ignored, so there is no accidental escalation to authenticated access. Use x-content-ai-mcp-api-key (not X-Api-Key) when you intend to search entitled content sources.
    note
    NOTE
    If you do not set the x-content-ai-mcp-routing header, the server asks for your environment (tier and bucket) in chat the first time you run a tool, then retries automatically.
    natural_language_search needs more than the other five tools: it requires cluster and namespace in routing, in addition to tier and bucket. If either is missing, the tool returns a missing_nls_routing error — reply with the missing value(s) and the server retries automatically.

  6. Click Add. The content-ai-mcp server now appears in your list of MCP Servers.

    Content AI MCP Server Listed {modal="regular"}

  7. Click New chat to start using the Content AI MCP Server.

    New Chat {modal="regular"}

Searching public content sources

Some Content AI content sources are marked public and can be searched anonymously, without an entitled identity. Public content sources are read-only and are intended for openly available, non-sensitive content. Anyone with the public API key can search them.

Use this server when:

  • You want to search openly available content (for example, public documentation or a demo catalog).
  • You do not need per-user, permission-scoped results.
  • You want a simple setup that only requires an API key.

To search public content sources, add the Content AI MCP Server in Coworker exactly as described in Add the Content AI MCP Server, but in the Add headers step add only the X-Api-Key header (plus an optional x-content-ai-mcp-routing header for the environment):

Header
Value
X-Api-Key
Your public API key.
x-content-ai-mcp-routing
(Optional) Environment/routing info, for example tier=publish,bucket=p12345-e67890.
IMPORTANT
When the X-Api-Key header is present, the server treats the request as anonymous and only searches public content sources. Any signed-in identity is ignored. Do not use X-Api-Key when you intend to search entitled content sources, use x-content-ai-mcp-api-key instead.

With the server connected, discover and search public content sources the same way as authenticated content sources.

  1. In Coworker, open a new chat and list the public content sources available to your API key:

    code language-text 
    List all Content AI content sources available in my environment.

    The server searches only public content sources for this request and returns matching results.

    The server returns the public sources matching your routing/environment.

You set up the Content AI MCP Server in Coworker and used it to search Content AI content sources. You discovered available content sources and searched them in natural language, with keyword, semantic, hybrid, and generative search. You also learned how to search public content sources anonymously with an API key, and how to reduce response size with field selection. You can use the same human-centric flow from Coworker to search and analyze Content AI content sources without switching to a UI or writing low-level search API code.

recommendation-more-help
experience-manager-learn-help-cloud-service