Every Drupal AI conversation now includes MCP, but which implementation should you actually use?
The Drupal AI Initiative describes two directions of work: “Inside AI,” where AI assists people within Drupal, and “Outside AI,” where external agents connect to Drupal and act on it. MCP sits at the center of both. Drupal can act as an MCP client, letting its own agents call external tools, and as an MCP server, letting assistants like Claude or Cursor work with a live site. In the server role there is no single standard yet. Plugin-based servers, Tool API-based servers, minimal “search and execute” designs and developer-focused knowledge servers all exist, with different strengths and levels of maturity.
This session is a practical field guide to that landscape. We’ll compare the options on tool granularity, where tools come from, authentication and permission models, maturity and security coverage, protocol currency and developer experience. A side-by-side demo will run the same task through different server styles so you can see the trade-offs in action. We’ll also be candid about the gaps: no settled default, limited audit and approval patterns for agent-made changes, open questions about editorial workflows, and the sustainability of community-maintained modules.
You’ll leave with:
- A clear map of the MCP options for Drupal today
- A decision framework for choosing between them
- A security checklist for letting AI agents act on your site
- Concrete ways to contribute to the gaps