DocumentationAEM as a Cloud ServiceUser Guide

Using MCP with AEM as a Cloud Service using-mcp-with-aem-as-a-cloud-service

Last update: Fri Mar 13 2026 00:00:00 GMT+0000 (Coordinated Universal Time)
  • Applies to:
  • Experience Manager as a Cloud Service

CREATED FOR:

  • User
  • Admin
  • Developer

Introduction introduction

Many Adobe Experience Manager (AEM) teams now work in Integrated Development Environments (IDEs) and chat-based applications such as Cursor, OpenAI ChatGPT, Anthropic Claude, and Microsoft Copilot Studio. These applications support the Model Context Protocol (MCP), which allows applications to expose back-end tools to large language models (LLMs) in a standardized way.

With AEM’s MCP integration, different personas can collaborate around the same content:

  • Developers can orchestrate content operations and workflows from their IDE or chat application
  • Practitioners and content architects can manage sites, content fragments, and assets with AI assistance while staying within AEM’s existing permission model.
IMPORTANT
For scenarios that modify or delete content, practitioners should use the AI Assistant interface rather than invoking MCP tools directly. The AEM Agents run by AI Assistant include built-in safeguards.

This article explains what AEM’s MCP functionality provides, which MCP applications are supported, how to configure it, and how to use it in practice.

Why MCP is Useful for AEM Customers why-mcp-is-useful-for-aem-customers

Modern IDE and chat applications use MCP as a way for an LLM to call tools exposed behind MCP servers. Customers can describe their intent in natural language instead of writing code against low-level API specifications. For example, a prompt like “update the hero banner for this campaign across all pages” lets the LLM invoke the appropriate MCP tools, which then interact with AEM’s APIs.

Key benefits include:

  • Natural-language interaction instead of API plumbing
    MCP tools describe what operations are available and how to call them. The LLM uses these schemas to decide which tools to invoke and with which parameters.
  • Consistent experience across applications
    The same AEM MCP tools can be used from multiple MCP-compatible applications, allowing teams to work where they are most productive while calling the same underlying AEM capabilities.
  • Security and governance preserved
    Requests to AEM MCP tools run under the authenticated user’s identity, and each tool enforces the user’s existing AEM permissions. AI-assisted operations follow the same access rules as manual work in AEM.

MCP Servers Provided by AEM mcp-servers-provided-by-aem

AEM exposes MCP servers as HTTP endpoints. The endpoints listed below are relative to:

https://mcp.adobeaemcloud.com/adobe/mcp/

MCP Servers mcp-servers

MCP Server
Endpoint
Description
Content
/content
All low-level content operations, including create, read, update, and delete (CRUD) for pages, fragments and assets.
Content (read-only)
/content-readonly
Read-only content operations (Get, List/Search) for pages, fragments, and assets.
Cloud Manager
/cloudmanager
Manage Cloud Manager entities including programs, environments, repositories and pipelines, which can also be triggered.

This MCP server is now in beta; to request access, email aemcs-mcp-feedback@adobe.com with a description of your use case.

The specific tools exposed by each MCP server may evolve over time. In practice, you can ask your MCP-enabled application to discover tools via a prompt such as:

"List all AEM MCP tools available from this server and describe what they do."

The MCP client uses the MCP protocol to retrieve the tool list and schemas, which the LLM can then use.

Reference the Content MCP Server Tutorial and Cloud Manager MCP Server Video for more information about their capabilities and how to use them.

Supported MCP Applications supported-mcp-applications

AEM’s MCP servers are designed to work with a defined set of MCP-compatible applications. Each application provides its own configuration experience, but the high-level steps are similar.

Chat Applications (Web & Desktop) chat-applications

  • Anthropic Claude
  • OpenAI ChatGPT

Developer Tools (IDE Extensions, Desktop Apps, CLIs) developer-tools

  • Anthropic Claude Code (CLI, JetBrains, VS Code, Cursor)
  • Augment Code (CLI, JetBrains, VS Code, Cursor)
  • Augment Indent Desktop App
  • Cline (JetBrains, VS Code, Cursor)
  • Cursor
  • GitHub Copilot (VS Code)
  • Kiro (Desktop App, CLI)
  • OpenAI Codex (Desktop App)
  • OpenAI Codex CLI
  • Windsurf

Enterprise Platforms enterprise-platforms

  • Microsoft Copilot Studio

Setup Overview setup-overview

Configuring MCP for AEM involves two main parts:

  1. Configuration in each MCP client application so that the application knows how to connect to the AEM MCP servers and perform OAuth login
  2. Select the MCP Server before starting to prompt, so that the MCP client knows to use it.

AEM Configuration aem-configuration

By default, the permissions that individual users have within AEM govern access to AEM’s MCP servers. When a user authenticates through an MCP client application, the MCP tools enforce the same access rules as manual operations in AEM. A user can only perform actions they are already authorized to perform.

Permitted MCP Client Applications permitted-mcp-client-applications

The following MCP client applications are permitted by default:

  • Anthropic Claude
  • Anthropic Claude Code
  • Augment Code
  • Augment Indent
  • Cline
  • Cursor
  • GitHub Copilot
  • Kiro
  • Microsoft Copilot Studio
  • OpenAI ChatGPT
  • OpenAI Codex
  • OpenAI Codex CLI
  • Windsurf

Restricting MCP Servers restricting-mcp-servers

All MCP servers are allowlisted by default. As an administrator, you have the option to restrict access to specific MCP servers at the organization, program, or environment level. This restriction gives you granular control over which MCP capabilities are available to users within your organization.

Managing MCP Client Access managing-mcp-client-access

Administrators can also disable access for specific MCP client applications if your organization’s policies require it. If you would like Adobe to enable support for additional MCP client products, send a link to the product website. If you need to allowlist a custom MCP client, reach out as well.

For all MCP server related requests, feel free to contact us at aemcs-mcp-feedback@adobe.com

MCP Client Application Configuration mcp-client-application-configuration

Each user performs this step, or an administrator of the MCP client application can perform it where supported. Configuration details vary slightly between applications. MCP clients are evolving rapidly, and support for remote MCP servers is being actively developed. You may need to enable Developer Mode to access the functionality for adding remote servers, but the general process is:

  1. Add one or more AEM MCP server URLs
    • Configure one or more MCP endpoints from the table above. For example:https://mcp.adobeaemcloud.com/adobe/mcp/content-readonly
  2. Trigger the connection
    • Save or activate the configuration so the MCP client application attempts to connect to the AEM MCP server
  3. Sign in with Adobe ID
    • When prompted, complete the Adobe login flow so the application can obtain OAuth tokens tied to your Adobe ID
  4. Verify discovered tools
    • Once authenticated, the application discovers MCP tools from the server. You can then start prompting the LLM to perform AEM operations.

Below are the supported applications, some of which link to step-by-step guides:

Chat Applications (Web & Desktop) setup-chat-applications

Developer Tools (IDE Extensions, Desktop Apps, CLIs) setup-developer-tools

  • Anthropic Claude Code (CLI, JetBrains, VS Code, Cursor)
  • Augment Code (CLI, JetBrains, VS Code, Cursor)
  • Augment Indent Desktop App
  • Cline (JetBrains, VS Code, Cursor)
  • Cursor
  • GitHub Copilot (VS Code)
  • Kiro (Desktop App, CLI)
  • OpenAI Codex (Desktop App)
  • OpenAI Codex CLI
  • Windsurf

Enterprise Platforms setup-enterprise-platforms

Authentication authentication

The Adobe-hosted MCP servers implement OAuth and are integrated with Adobe’s identity system.

  • When an MCP client application connects to an AEM MCP server, users see an Adobe login dialog and authenticate with their Adobe ID
  • After successful login, the system verifies that the MCP client application is permitted in your organization and that the requested MCP server is allowed. If either check fails, an error message is displayed.

MCP Client not permitted error

  • Once verified, the MCP server issues tokens that the application uses for subsequent tool calls
  • MCP tools respect the user’s AEM permissions. Only users who have permission to modify a content fragment in AEM can modify it via MCP.

This approach ensures that AI-assisted operations comply with your existing AEM security and governance model.

Using MCP with AEM using-mcp-with-aem

Once AEM and your MCP client applications are configured, you can work in your application of choice and prompt the LLM to perform AEM operations. The LLM reads the MCP tool schemas, chooses which tools to call, and sequences them as needed to fulfill your request.

IMPORTANT
Prompts that contain multiple steps or target different content types, such as images and text, work best with a thinking model. Enable a thinking model or select the Thinking option in your MCP client instead of relying on Auto mode.

Example Usecases example-usecases

Some representative scenarios include:

  • Environment discovery

    • List environments and licenses to decide where to run a workflow.

  • Sites management

    • List sites
    • Create, read, update, and delete pages and page content.

  • Content Fragment management

    • Search for content fragments
    • Create new fragments
    • Update existing fragments when campaign messaging changes.

  • Assets management

    • Import assets
    • Find existing assets
    • Publish assets.

Example Workflows example-workflows

The following examples illustrate how an LLM might chain MCP tools together.

  1. Working with a content fragment referenced by a page

    • Get page content – Call a tool such as get-aem-page-content to retrieve the page and locate the fragmentPath property.
    • Resolve the fragment path – Use resolve_fragment_path to convert the path to a UUID.
    • Fetch fragment data – Call get_fragment to retrieve the current fields.
    • Update the fragment – Call patch_fragment to apply changes to the fragment content.

  2. Creating new content based on a model

    • Discover models – Use list_models to see which content fragment models are available.
    • Inspect a model – Use get_model to understand the model’s field schema.
    • Create content – Use create_fragment to create a new fragment with values derived from your prompt.

  3. Safely updating existing content

    • Read current data – Use get_fragment to retrieve the existing data and an ETag.
    • Apply a JSON Patch – Use patch_fragment with the ETag and a JSON Patch document to update the fragment, supporting optimistic concurrency.

From the user’s perspective, these workflows can be initiated with prompts such as:

"Create a new content fragment for the spring campaign based on our hero banner model and fill in its fields from this brief."

The LLM chooses and coordinates the necessary MCP tools automatically.

Expectation Management expectation-management

When working with LLMs through MCP, keep the following in mind:

  • Highly capable but not infallible
    LLMs can accomplish complex tasks but are prone to occasional errors. The same prompt may yield slightly different results or presentations without an obvious reason. Always review outputs before applying changes to production content.

  • Evolving capabilities
    LLM models are continuously improving. Over time, they become smarter at discovering new ways to combine MCP tools to achieve your goals. A task that required multiple prompts today may work seamlessly with a single prompt tomorrow.

  • Human oversight is essential:
    Think of the LLM as a knowledgeable assistant that needs supervision. It has broad knowledge and can devise creative solutions, but it benefits from your guidance and review. Verify results, especially for critical operations, and provide feedback when the output does not match your expectations.

  • Be cautious with auto-acknowledging tool executions
    Some MCP client applications, such as Claude, offer the option to auto-acknowledge tool executions requested by the LLM. While this option can be convenient for read-only operations like searching or retrieving content, exercise caution with tools that update or delete content. Review each tool execution request before confirming actions that modify your AEM environment.

Limitations limitations

AEM currently supports configuring MCP servers in the applications listed under Supported MCP Applications.

If you would like to use a different MCP client application, feel free to reach out at aemcs-mcp-feedback@adobe.com to request support for additional clients or to allowlist a custom one.

recommendation-more-help
fbcff2a9-b6fe-4574-b04a-21e75df764ab