IBM Content Services MCP Server

Overview

The IBM Content Services MCP Server provides a standardized interface that enables IBM FileNet Content Manager (FNCM) capabilities to be used by AI models. The server is available in three specialized configurations to support different workflows:

• Core Server: Comprehensive document and content management operations

• Property Extraction and Classification Server (Preview): AI-powered document analysis and classification

• Legal Hold Server (Preview): Legal hold and compliance management

Each server can be deployed independently or in combination to support your specific use cases.

Server Types and Capabilities

Core Server (core-cs-mcp-server)

Purpose: Comprehensive document and content management operations

Use Cases:

• General document lifecycle management (create, update, check-in, check-out, delete)

• Folder operations and document filing

• Search and retrieval across the repository

• Metadata and class management

• Document annotations

• Resources support for exposing documents as LLM context

Property Extraction and Classification Server (property-extraction-and-classification-cs-mcp-server) - Preview

Purpose: AI-powered document analysis and classification

Dependencies: Requires Core Server for document update operations (e.g., update_document_properties)

Use Cases:

• Automated property extraction from document content

• Document classification and reclassification workflows

• Content-based metadata enrichment

Legal Hold Server (legal-hold-cs-mcp-server) - Preview

Purpose: Legal hold and compliance management

Use Cases:

• Creating and managing legal holds

• Placing documents and objects under hold

• Tracking held objects

• Compliance and litigation support workflows

Tools List

Core Server Tools

The Core Server provides 28 tools organized into the following categories:

Document Management (11 tools)

• get_document_versions: Retrieves a document's version history, including major and minor version numbers and document IDs for each version.

• get_document_text_extract: Extracts text content from a document by retrieving its text extract annotations. If multiple text extracts are found, they are concatenated. Note: This functionality requires the Persistent Text Extract add-on to be installed in your object store. See the Prerequisites section for more details.

• create_document: Creates a new document in the content repository with specified properties. Can upload files as the document's content if file paths are provided. Requires first calling determine_class and get_class_property_descriptions.

• update_document_properties: Updates an existing document's properties without changing its class. Requires first calling get_class_property_descriptions to get valid properties for the document's current class.

• update_document_class: Changes a document's class in the content repository. WARNING: Changing a document's class can result in loss of properties if the new class does not have the same properties as the old class. Requires first calling determine_class to get the new class_identifier.

• checkin_document: Checks in a document that was previously checked out. Can upload new content files during check-in if file paths are provided.

• checkout_document: Checks out a document for editing. Can download the document content to a specified folder path if provided.

• cancel_document_checkout: Cancels a document checkout in the content repository, releasing the reservation.

• get_document_properties: Retrieves a document from the content repository by ID or path, returning the document object with its properties.

• delete_document_version: Deletes a specific document version in the content repository using its document ID.

• delete_version_series: Deletes an entire version series (all versions of a document) in the content repository using the version series ID.

Folder Management (7 tools)

• create_folder: Creates a new folder in the content repository with specified name, parent folder, and optional class identifier.

• delete_folder: Deletes a folder from the repository using its ID or path.

• unfile_document: Removes a document from a folder without deleting the document itself.

• file_document: Files a document into a folder.

• update_folder: Updates an existing folder's properties. Requires first calling determine_class and get_class_property_descriptions.

• get_folder_documents: Get documents contained in a folder.

• get_folder_detail: Retrieves detailed information about a folder.

Class/Metadata Management (3 tools)

• list_root_classes: Lists root classes.

• determine_class: Determines the appropriate class based on the available classes and the content of the user's message or context document.

• get_class_property_descriptions: Retrieves detailed descriptions of all properties for a specified class.

Search (4 tools)

• get_searchable_property_descriptions: Retrieves descriptions of properties that can be used in search operations.

• repository_object_search: Searches for repository objects based on specified criteria.

• lookup_documents_by_name: Searches for documents by matching keywords against document names. Returns a ranked list of matching documents with confidence scores. Useful when you know part of a document's name but not its exact ID or path.

• lookup_documents_by_path: Searches for documents based on their location in the folder hierarchy. Matches keywords against folder names and document containment names at each path level. Particularly useful when the user describes a document using path separators (e.g., "/Folder1/Subfolder/document").

Annotations (2 tools)

• get_document_annotations: Retrieves all annotations associated with a document, including their IDs, names, descriptive text, and content elements.

• get_annotation: Retrieves a specific annotation by its ID.

Custom Objects (1 tool)

• get_custom_object: Retrieves a custom object from the repository by ID or path.

Property Extraction and Classification Server Tools (Preview)

The Property Extraction and Classification Server provides 2 specialized tools for AI-powered document workflows:

• property_extraction: Extracts document class, properties metadata, and text content for AI-based property value extraction. This tool determines the document's class, fetches class metadata to identify all available properties (filtering out system and hidden properties), and retrieves the document's text extract content.

• list_all_classes: Lists all available classes for a specific root class type. Essential for document reclassification workflows where you need to match document content to the most appropriate class.

Legal Hold Server Tools (Preview)

The Legal Hold Server provides 6 tools for legal compliance management:

• create_hold: Creates a new legal hold with a specified display name.

• delete_hold: Removes a legal hold and releases all held objects.

• add_object_to_hold: Places an object (document, folder, etc.) under a legal hold.

• delete_object_from_hold: Removes an object from a legal hold without deleting the hold itself.

• get_held_objects_for_hold: Lists all objects currently under a specific legal hold.

• get_holds_by_name: Searches for legal holds by their display name.

Resources (Core Server Only)

What are Resources?

Resources provide read-only access to document content for LLM context. Documents in a configured folder are automatically exposed as MCP resources, allowing AI models to reference them during conversations without explicit tool calls.

Note: Resources functionality requires the Persistent Text Extract Add-on to be installed in your object store to retrieve document content. See the Prerequisites section for installation details.

Configuration

Set the RESOURCES_FOLDER environment variable to specify the folder path in your object store where resource documents should be uploaded:

The RESOURCES_FOLDER value is the path of the folder in your FileNet object store where you should upload your resource documents. The server will automatically discover and register all documents in this folder.

Documents in this folder will be:

• Automatically registered as resources when the Core server starts

• Available to the LLM with URIs following the pattern: ibm-cs://{object_store}/documents/{folder_path}/{document_name}

• Displayed with names in the format: [IBM CS] {document_name}

Security Guidelines

Access Control: The resources folder should only be modifiable by administrative users to prevent tampering with content that will be used by AI models. Configure appropriate folder permissions in your FileNet object store to:

• Restrict write/modify access to administrators only

• Allow read access for the MCP server business user

• Prevent unauthorized users from adding, modifying, or deleting resource documents

Use Cases

Resources are ideal for providing AI models with reference documentation:

• Policy documents for compliance guidance

• Classification guidelines for document categorization

• Standard operating procedures

• Regulatory requirements

• Best practices documentation

Example

If you set RESOURCES_FOLDER=/policies and have documents in that folder:

• /policies/data_classification_policy.txt

• /policies/retention_policy.txt

These documents will be available as resources that the AI can reference when answering questions or making decisions.

Tested Environments

The Content Services MCP Servers have been tested with the following MCP client and LLM combinations:

• Claude Desktop: Sonnet 4.5, 4, 3.5 and Haiku 4.5

• Watsonx Orchestrate: Llama-3-2-90b-vision-instruct

While other MCP client and LLM combinations have not been tested, they may work with this server. We encourage you to experiment and validate for yourself.

For setup instructions with additional MCP clients, see:

• Bob-IDE MCP Server Setup

• VS Code Copilot MCP Server Setup

MCP Client Limitations

Some MCP clients have limitations that affect which tools can be used. The following table shows known compatibility issues:

Note: These limitations are due to the MCP client's input handling capabilities, not the MCP server itself.

Setup and Configuration

Prerequisites

• Python 3.13+

• uv

  • on macOS: brew install uv

  • on Windows: see link above

• Access to a FileNet CPE server with Content Services GraphQL API (CS-GQL) installed

• Persistent Text Extract Add-on must be installed in your object store if you want to use document content retrieval functionality

  • This add-on enables the extraction and storage of text content from documents

  • Required for: Core Server's get_document_text_extract tool, Resources functionality, and Property Extraction Server's property_extraction tool

  • Without this add-on, these features will not return document content

  • For installation instructions, refer to the IBM Documentation on Installing the Persistent Text Add-on

Configuration

The Content Services MCP Servers require several environment variables to connect to your FileNet CPE server:

Required Environment Variables

Optional Environment Variables

Cloud Pak for Business Automation Environment Variables

SSL Configuration Best Practices

For SSL configuration (SSL_ENABLED, TOKEN_SSL_ENABLED, ZENIAM_ZEN_SSL_ENABLED, and ZENIAM_IAM_SSL_ENABLED), you have three options:

1. Use System Certificates (Recommended for Production): Set to true to use your system's certificate store.

2. Provide Custom Certificate Path: Set to the file path of your certificate (e.g., /path/to/certificate.pem).

3. Disable SSL Verification (Not Recommended for Production): Set to false to disable SSL verification.

Security Warning: Disabling SSL verification (false) should only be used in testing environments. For production deployments, always use proper certificate validation to ensure secure communications.

Authentication Methods

The server supports three authentication methods:

Basic Authentication

Set the following environment variables:

OAuth Authentication

Set the following environment variables:

Zen/IAM Authentication

An example of ZEN/IAM environment variables when using USER/PASSWORD and SSL to all external servers

Integration with MCP Clients/Agent Frameworks

Claude Desktop Configuration

1. Open Claude Desktop Settings:

   • On macOS, click the Claude menu in the top menu bar and select Settings.

   • On Windows, access Settings from the Claude application.

2. Navigate to the Developer tab and click Edit Config:

   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

   • Windows: %APPDATA%\Claude\claude_desktop_config.json

3. Add one of the following configuration examples to the claude_desktop_config.json file:

   Option 1: Using local installation (if you've cloned the repository)

   { "mcpServers": { "core-cs-mcp-server": { "command": "/path/to/your/uvx", "args": [ "--from", "/path/to/your/cs-mcp-server", "core-cs-mcp-server" ], "env": { "USERNAME": "your_username", "PASSWORD": "your_password", "SERVER_URL": "https://your-graphql-server/content-services-graphql/graphql", "OBJECT_STORE": "your_object_store" } } } }

   Option 2: Installing directly from GitHub (recommended)

   { "mcpServers": { "core-cs-mcp-server": { "command": "uvx", "args": [ "--from", "git+https://github.com/ibm-ecm/ibm-content-services-mcp-server", "core-cs-mcp-server" ], "env": { "USERNAME": "your_username", "PASSWORD": "your_password", "SERVER_URL": "https://your-graphql-server/content-services-graphql/graphql", "OBJECT_STORE": "your_object_store" } } } }

   Option 3: Multi-Server Setup (Core + Property Extraction + Legal Hold)

   { "mcpServers": { "core-cs-mcp-server": { "command": "uvx", "args": [ "--from", "git+https://github.com/ibm-ecm/ibm-content-services-mcp-server", "core-cs-mcp-server" ], "env": { "USERNAME": "your_username", "PASSWORD": "your_password", "SERVER_URL": "https://your-graphql-server/content-services-graphql/graphql", "OBJECT_STORE": "your_object_store" } }, "property-extraction-cs-mcp-server": { "command": "uvx", "args": [ "--from", "git+https://github.com/ibm-ecm/ibm-content-services-mcp-server", "property-extraction-and-classification-cs-mcp-server" ], "env": { "USERNAME": "your_username", "PASSWORD": "your_password", "SERVER_URL": "https://your-graphql-server/content-services-graphql/graphql", "OBJECT_STORE": "your_object_store" } }, "legal-hold-cs-mcp-server": { "command": "uvx", "args": [ "--from", "git+https://github.com/ibm-ecm/ibm-content-services-mcp-server", "legal-hold-cs-mcp-server" ], "env": { "USERNAME": "your_username", "PASSWORD": "your_password", "SERVER_URL": "https://your-graphql-server/content-services-graphql/graphql", "OBJECT_STORE": "your_object_store" } } } }

4. Restart Claude Desktop:

   • Simply closing the window is not enough, Claude Desktop must be stopped and restarted:

     • on macOS: Claude > Quit

     • on Windows: File > Exit

5. Check Available Tools:

   • To see all the available tools in Claude Desktop, proceed as follows:

     • first click the settings icon, and you should see:

     • then click core-cs-mcp-server, and you should see all your tools:

Note: The JSON configuration examples above show only the minimum required environment variables. For a complete list of all possible configuration options, refer to the Environment Variables tables above.

Watson Orchestrate (WxO) Configuration

This section explains how to augment IBM watsonx Orchestrate with the Content Services MCP Servers, enabling watsonx Orchestrate to interact with IBM FileNet Content Management during user interactions in a chat.

You can configure one or multiple servers depending on your needs:

• Core Server: For general document management operations

• Property Extraction and Classification Server: For AI-powered document analysis (requires Core Server for document updates)

• Legal Hold Server: For legal compliance workflows

The configuration steps below use the Core Server as an example, but the same process applies to other servers by changing the server name and install command.

Configuration

1. Configure Connection Variables

For SaaS or on-premises offering (UI):

• Click the main menu icon

• Navigate to Manage > Connections

• Click Add New Connection

• Enter connection ID and display name

• Click Next

• You will now configure draft connection details (test environment)

  • Select authentication type dropdown to be Key value pair

  • Enter each required variable:

    • SERVER_URL: Your Content Services GraphQL API endpoint URL

    • USERNAME: Authentication username

    • PASSWORD: Authentication password

    • OBJECT_STORE: Object store identifier

  • Enter any optional variables as needed (e.g., SSL_ENABLED, TOKEN_REFRESH, etc.)

  • Click Next when done

• Now you will enter your live connection environment variables

  • Select authentication type dropdown to be Key value pair

  • Enter the same required variables as above

  • Enter any optional variables as needed

  • Select the preferred credential type

  • Click Add Connection

For ADK (Application Development Kit):

For creating connections using the ADK CLI, please refer to the official documentation.

2. Create an agent

• Click the main menu icon

• Navigate to Build > Agent Builder

  

• Navigate to All agents

• Click Create agent + to add a new agent

  

• Choose Create from scratch

• Enter a Name (e.g., Core Content Services Agent)

• Enter a Description (e.g., This agent enables interaction with FileNet Content Management.)

• Click Create

  

3. Augment the agent with the Core Content Services MCP Server

• Navigate to the Toolset section, click Add tool +

  

• Click Import

  

• Click Import from MCP server

  

• Click Add MCP server

  

• Enter a Server name without any space characters (e.g., core-cs-mcp-server)

• Optionally enter a Description (e.g., This MCP Server connects to FileNet Content Platform Engine, enabling content management operations.)

• Enter an Install command:

  uvx --from git+https://github.com/ibm-ecm/ibm-content-services-mcp-server core-cs-mcp-server

  For other servers, use these install commands:

  • Property Extraction and Classification Server:

    uvx --from git+https://github.com/ibm-ecm/ibm-content-services-mcp-server property-extraction-and-classification-cs-mcp-server

  • Legal Hold Server:

    uvx --from git+https://github.com/ibm-ecm/ibm-content-services-mcp-server legal-hold-cs-mcp-server

• Click Connect

• If you see "Connection successful", click Done

  

• Set the Activation toggle to On for the tools you want to enable

  

• Associate your previously created connection with this agent

4. Deploy the agent

• Click Deploy

  

• In the popup, Click Deploy again

5. Let the agent be used in chats

• Click the main menu icon

• Navigate to Chat

• Click the newly created agent

  

Example Workflow

Once configured, you can interact with your FileNet repository through natural language in watsonx Orchestrate chats, depending on which tool you've enabled. For example:

• "Find all documents containing the pdf in its document title"

• "Create a new folder called Project Z"

  

Click Show Reasoning in any response to see the details of the operations performed.

Usage

Running Servers Directly

If you have a local copy of the repository, you can run any server directly with environment variables:

Core Server:

Property Extraction and Classification Server:

Legal Hold Server:

Integration with AI Agents

The Content Services MCP Servers can be integrated with AI Agents that support the MCP protocol. Depending on which server(s) you deploy, the AI Agent can:

Core Server capabilities:

1. Access and retrieve document properties

2. Extract text from documents

3. Create, update, check-in, and check-out documents

4. Manage folders and document classifications

5. Execute searches

6. Get document annotations

7. Access resources for LLM context

Property Extraction and Classification Server capabilities:

1. Extract property values from document content using AI

2. List and match document classes for reclassification

Legal Hold Server capabilities:

1. Create and manage legal holds

2. Place objects under hold

3. Track and query held objects

Example Workflows

Core Server Workflow: Search and Document Management

1. Search and Discovery:

   • Users typically start with descriptive information (name, content, keywords) rather than IDs

   • The AI Agent first uses search tools to locate relevant objects:

     • get_searchable_property_descriptions to discover valid search properties

     • repository_object_search for property-based searches

   • Search results include object IDs needed for subsequent operations

2. Document Retrieval:

   • Once an object ID is obtained through search, the AI Agent can retrieve:

     • Document properties using the ID

     • Version history

     • Text content (requires Persistent Text Extract Add-on to be installed)

     • Annotations

3. Document Creation: Users can ask the AI Agent to create new documents with specific properties and content.

4. Document Update:

   • After identifying a document through search, the AI Agent can:

     • Check out the document using its ID

     • Update properties or content

     • Check the document back in

5. Folder Operations:

   • Folders can be identified by path or by ID from search results

   • Documents can be filed/unfiled using both document and folder IDs

Property Extraction and Classification Workflow

Requires: Core Server (for document updates)

1. Property Extraction:

   • Use property_extraction tool with a document ID

   • The tool returns:

     • Document class information

     • All available properties for that class (excluding system/hidden properties)

     • Document text content

   • AI analyzes the text and extracts appropriate property values

   • Use Core Server's update_document_properties to save the extracted values

2. Document Classification:

   • Use list_all_classes to get all available document classes

   • AI analyzes document content and matches it to the most appropriate class

   • Use Core Server's update_document_class to reclassify the document

Legal Hold Workflow

1. Creating a Hold:

   • Use create_hold with a descriptive name (e.g., "Litigation ABC vs XYZ")

   • Returns the hold object with its ID

2. Placing Objects Under Hold:

   • Identify documents using Core Server search tools (if needed)

   • Use add_object_to_hold with the hold ID and object ID

   • Repeat for all relevant documents/objects

3. Managing Holds:

   • Use get_holds_by_name to find holds by name

   • Use get_held_objects_for_hold to see all objects under a specific hold

   • Use delete_object_from_hold to release specific objects

   • Use delete_hold to remove the hold entirely (releases all objects)

Note: Most operations that modify or access specific objects require an object ID, which is typically obtained through a search operation first. This workflow pattern ensures users can work with objects by their meaningful attributes rather than requiring them to know technical identifiers upfront.

License

See the LICENSE file for details.