In this hands-on activity, participants will use Claude Desktop to manage a local WordPress site through an AI agent via natural language conversation. Without needing advanced prerequisite knowledge, attendees will build their environment using a real local WordPress server (such as MAMP, Studio by WordPress.com, or DevKinsta) and install three core open-source plugins: the MCP Adapter, WP Abilities API Demo, and WP MCP Server Demo. Participants will gain hands-on experience configuring non-plain permalinks, generating a WordPress Application Password, and modifying the Claude Desktop configuration file using the @automattic/mcp-wordpress-remote package. The final part of the session entails a live execution workflow where developers will prompt Claude to query active plugins, check the system debug log, generate a draft post directly on their local database, and test custom free-form natural language prompts.
Learning Objectives
By the end of this activity, participants will be able to:
- Identify MCP Mechanics: Define what the Model Context Protocol (MCP) is and describe how it exposes site capabilities as tools to allow external AI assistants to interact with WordPress.
- Differentiate AI Architectures: Explain the difference between native dashboard-bound AI features that push content into the editor and external AI agents managing WordPress from the outside via the Abilities API.
- Configure WordPress Plugins: Install and activate the MCP Adapter, WP Abilities API Demo, and WP MCP Server Demo plugins on a local site via the admin dashboard zip upload method.
- Establish Secure API Connections: Generate a 24-character WordPress Application Password and integrate it alongside the site URL and admin username into the Claude Desktop JSON configuration file using the
npxcommand. - Execute Site Tasks via AI: Utilize natural language commands inside Claude Desktop to fetch active plugin lists, inspect the WordPress debug log, and create staged draft posts.
- Evaluate AI Autonomy: Assess which site management tasks are safe to delegate to an automated AI agent and determine which structural actions necessitate staging or human review before final execution.