drupal-api

Drupal MCP

This plugin exposes a Drupal site's core JSON:API as MCP tools. It works on any Drupal 10/11 site with the jsonapi core module enabled.

Tools

• drupal_list_nodes — list nodes of a bundle (filter / sort / paginate).
• drupal_get_node — fetch one node by UUID.
• drupal_create_node — create a new node.
• drupal_update_node — patch attributes or relationships.
• drupal_delete_node — delete by UUID.
• drupal_list_taxonomy_terms — list terms in a vocabulary.
• drupal_list_users — list users.
• drupal_query_jsonapi — escape hatch for arbitrary GETs.

Conventions

• All identifiers are UUIDs (JSON:API canonical), not numeric nids. Use drupal_list_nodes first to find the UUID.
• Bundle = content type machine name (e.g. article, page).
• attributes is a flat object of JSON:API field names → values, e.g. { title: '...', body: { value: '...', format: 'basic_html' }, status: true }.
• relationships follow JSON:API shape: { field_image: { data: { type: 'file--file', id: 'uuid' } } }.
• Filter shorthand: { field_category: '1' } becomes filter[field_category]=1. Operators: { field_category: { value: '1', operator: '=' } }.

Auth + permissions

The MCP runs as a configured Drupal user. Permissions on that user's role decide what is reachable. For least privilege, create a mcp_bot user with a dedicated role that only grants the bundles/operations you actually need. JSON:API ignores the access_check flag — Drupal field access still applies.

Pre-flight

Before any write, confirm with the user:

1. Which bundle / vocabulary?
2. Which environment (prod vs dev)?
3. Show the full payload before issuing the call.

Deletion is irreversible. Always read first, then ask, then delete.

Common patterns

Find the latest published articles:

drupal_list_nodes(bundle='article', filter={status: 1}, sort='-created', limit=5)

Update an article body:

drupal_update_node(bundle='article', uuid='<uuid>', attributes={ body: { value: '...', format: 'basic_html' } })

Lookup taxonomy term ID for filter chains:

drupal_list_taxonomy_terms(vocabulary='category')
# then use the returned UUID in a node filter