v2.19.4

What's Changed

The field-test follow-ups from #336, contributed by the field-tester. The headline is an upstream bug: GET /databases can drop whole database types, and list_databases now repairs that from /resources.

Fixed

• list_databases no longer silently drops whole database types (#336, item 1). Coolify keeps a per-type id sequence and GET /databases merges the per-type collections keyed on that id, so two types sharing ids shadow each other — verified live upstream: 2 standalone Postgres + 2 standalone Dragonfly, both pairs ids 1 and 2, returned only the Dragonflys. /resources reports every database correctly, so listDatabases now fetches both and merges the rows /databases dropped back in by uuid (every standalone-* resource type is a database). The merge is a no-op when /databases is complete, and if /resources fails the plain /databases result is returned unchanged. get_infrastructure_overview database counts and find_issues database coverage are repaired by the same merge, since both go through listDatabases.
• environments get now honours its own schema (#336, item 2). The schema has always marked name optional, but the handler rejected with Error: name required. It now defaults to the sole environment when the project has exactly one; with several (or none) it explains what exists instead of a bare rejection.
• diagnose_app env-var counts no longer read as a bug (#336, item 3). Coolify auto-creates a preview twin for every production application env var and the API returns both scopes merged, so the raw count doubles. The diagnostic now surfaces is_preview per variable plus distinct_keys, production_count and preview_count alongside the unchanged raw count. No row is ever deduped — deleting a twin destroys preview config.
• Exact matches win name resolution (#336, item 4). diagnose_app resolved "api.example.com" to a multi-app disambiguation error whenever another app's FQDN contained it as a substring. An exact name or FQDN match (scheme and trailing slash ignored, each entry of a comma-separated fqdn compared) now wins outright before substring matching runs; same for diagnose_server names and IPs, where "10.0.0.5" is a substring of "10.0.0.50".

Added

• find_issues folds in signals it already had in hand (#336, item 5). An app in running:unknown and an available proxy patch upgrade used to produce zero findings. Issues now carry a severity (critical | warning): resources running with unknown health warn (running container, no working healthcheck — a failure would go undetected), and servers with an available Traefik update warn with the version pair (traefik_outdated_info only exists on GET /servers/{uuid}, so each listed server is fetched individually; shape verified live against v4.3.7). The unhealthy_*/unreachable_servers summary counts keep their pre-existing critical-only meaning; warnings are counted separately in summary.warnings.

---

📦 View on npm

🖱️ One-click Claude Desktop install: download coolify-mcp.mcpb below and drag it into Settings → Extensions.