mcp-adapter

The MCP Adapter is a WordPress-focused MCP server implementation that exposes WordPress abilities as MCP tools, resources, and prompts. It bridges WordPress' Abilities API with the Model Context Protocol so AI agents can discover, query, and execute WordPress capabilities in a structured MCP format. The adapter supports multiple transport methods, including HTTP for remote, API-driven interactions and STDIO for local development or CLI integrations. It also provides extensible hooks for custom error handling, observability, and per-server configurations. With this setup, you can discover built-in WordPress abilities, convert them into MCP components, and use them from agents that understand MCP prompts, tools, and resources.

Prerequisites:

• Docker installed and running
• Access to a WordPress instance (self-hosted or hosted) with any required WordPress abilities API enabled
• Optional: PHP and Composer if you prefer a local, non-Docker development environment

Installation steps (Docker-based):

1. Pull the adapter image (as needed) or build from Dockerfile if provided: docker pull wordpress/mcp-adapter

2. Create a configuration file or environment variables to configure the adapter for your WordPress instance. Example envs to set:

   • WORDPRESS_URL: https://your-wordpress-site.example
   • WORDPRESS_API_KEY: your-api-key (if the WordPress Abilities API requires it)
   • MCP_TRANSPORTS: http,stdio (or your desired transports)
   • MCP_LOG_LEVEL: info

3. Run the adapter using Docker (example based on the provided mcp_config): docker run -i
   -e WORDPRESS_URL="https://your-wordpress-site.example"
   -e WORDPRESS_API_KEY="your-api-key"
   -e MCP_TRANSPORTS="http,stdio"
   -e MCP_LOG_LEVEL="info"
   wordpress/mcp-adapter

4. Verify the server starts and registers with the MCP registry as per your environment. If you expose HTTP, ensure the host/port are accessible to your MCP clients. If you use STDIO transport, you can wire it into CLI workflows or local development pipelines.

5. Optional: integrate with your existing WordPress setup (e.g., enable additional abilities, adjust permissions, and configure per-server handlers).

If you prefer a local PHP/Composer setup, you would clone the repository, install PHP dependencies, and run the built-in server or a CLI bridge as described in the repository’s developer docs.

Tips and common questions:

• The adapter supports multiple transport methods concurrently; you can configure HTTP for remote AI agents and STDIO for local testing in the same server instance.
• Ensure proper permissions are set for the exposed MCP endpoints and that WordPress abilities you want to expose are enabled in your WordPress setup.
• If you encounter transport or serialization issues, verify the MCP version compatibility between your agent and the adapter (the README indicates MCP 2025-06-18 alignment).
• Use the per-server configuration capabilities to tailor error handling, observability, and validation rules for each MCP server you deploy.
• For observability, consider wiring in your existing monitoring stack via the McpObservabilityHandlerInterface implementations described in the architecture notes.
• When using HTTP transport, ensure CORS, authentication, and network policies permit the agent to reach the adapter endpoint.
• If you are developing locally, STDIO transport can be a fast iteration path to test abilities, prompts, and prompts-driven workflows before exposing an HTTP API.