mcp

IBM MCP is a curated collection of MCP servers that enables AI agents to securely interact with a variety of IBM services and data sources. This bundle includes servers for workflow automation, content management, decision services, data access, and more, such as IBM MQ access, Business Automation workflows, FileNet content interactions, decision intelligence, and data lake integrations. Each server exposes a standardized MCP interface so you can query, invoke actions, and retrieve results from the underlying IBM services through natural language or structured requests. To get started, install the MCP tooling (often via the VS Code extension or through the command palette) and launch the specific IBM MCP server you want to work with. The README links indicate how you can install or run individual servers in a local environment or via VS Code configurations, with environment variables to customize endpoints, credentials, and targets. When you run a server, you’ll typically connect your AI agent to its MCP endpoint and begin issuing domain-specific requests (e.g., querying a document repository, invoking a decision service, or managing a workflow). The tools available span enterprise content management (Content Services), decision orchestration (Decision Intelligence/ODM), workflow automation (BAW), and data access (watsonx.data, OpenPages, Astra DB, Docling, etc.). Use the provided env placeholders to configure authentication and endpoints as you wire up your AI agents to the IBM services you rely on.

Prerequisites:

• Docker installed and running on your machine (for the provided docker-based setup).
• Optional: VS Code with the MCP extension for streamlined installation and in-editor execution (see the README for install badges and commands).

Step-by-step:

1. Ensure your environment meets prerequisites (Docker installed, credentials available for the IBM services you intend to use).
2. Pull or run the IBM MCP server collection via Docker:
   • docker pull ibm-mcp-server:latest
   • docker run -i --name ibm-mcp-server ibm-mcp-server:latest
3. If you prefer local development via VS Code, follow the VS Code install links in the README for each server you want to enable. The links show commands like:
   • uvx (Python/uv) for stdio-based servers
   • npx for quick npm-based server launches
   • Specific env examples to populate endpoints, tokens, and credentials
4. Configure environment variables for your target IBM services as shown in the README examples (ENDPOINT, USERNAME, PASSWORD, API keys, etc.).
5. Connect your MCP client (AI agent) to the server’s MCP endpoint as described in your chosen server’s docs and begin sending requests.

Notes:

• If you’re using the VS Code workflow, you’ll typically use a config snippet that points to the server type (uvx, npx, etc.) with your endpoint and credentials embedded in env vars.
• For production setups, consider container orchestration and proper secret management for credentials.

Tips and caveats:

• Some IBM MCP servers expose REST-like or standardized MCP endpoints; ensure your agent uses the correct request format for each server (endpoints and actions vary by service).
• Environment variables are often placeholders; replace with real credentials or use a secret store when deploying.
• If you encounter connectivity issues, verify Docker is running, the container is healthy, and that the endpoint URLs are reachable from the host.
• The collection includes a mix of local and cloud-hosted offerings; choose the deployment mode (local dev vs. production) accordingly, and consult each server’s specific docs for authentication and endpoint details.
• When using the VS Code install links, the command payloads typically include --from, --from and --name fields to pull the server from GitHub and configure the runtime (uvx or npx).