Enhancement: Replace hardcoded Container Apps templates with MCP tool (apply #1949 pattern)

[paulyuk]

DRIs: @paulyuk (assignee), @simonjj (co-DRI — see comment, external contributor)

---

Context

@manvkaur's PR #1949 replaced ~9,000 lines of hardcoded Azure Functions templates with dynamic MCP tool calls (functions_template_get), with a 3-tier fallback (CDN manifest → GitHub source → known-good Azure-Samples repos). This dramatically reduced context size and maintenance burden.

Container Apps should adopt the same pattern.

Scope

The Container Apps templates introduced in #1610 / PR #1636 currently hard-code:

• 15 files under plugin/skills/azure-prepare/references/services/container-apps/templates/
• ~2,033 lines covering 6 base templates (web-app, api, microservice, worker, job, functions-on-aca) and 7 recipes (acr, cosmos, dapr, postgres, redis, servicebus, composition)

These could be replaced with an analogous container_apps_template_get (or similar) MCP tool that fetches templates dynamically from Azure-Samples repos.

Proposed approach

1. Source of truth — identify or create AZD templates in Azure-Samples for each base template × recipe combination (Container Apps already has rich AZD templates: container-apps-azd-templates)
2. MCP tool — extend the Azure MCP server with container_apps_template_get mirroring functions_template_get semantics
3. Fallback chain — same 3-tier pattern: CDN manifest → GitHub source → known-good repos
4. Skill refactor — delete hardcoded base template files; keep orchestration (composition.md, selection.md, recipes/README.md) with MCP tool guidance like Functions
5. Recipe-specific work — Container Apps recipes (Dapr components, KEDA scalers) are more layered than Functions bindings. Keep recipe READMEs that document the additive overlays; let the MCP tool deliver the base + Dapr/KEDA wiring patterns
6. Eval coverage — add cross-language eval summaries proving end-to-end deployment

Why this isn't blocking PR #1636

PR #1636 ships proven, manually-curated templates that work today. The MCP tool approach requires:

• New MCP server endpoint
• AZD template inventory mapped to base+recipe combinations
• Cross-team coordination

We don't want to delay shipping Container Apps Develop equity for an optimization. Ship #1636 first, then layer this enhancement on top.

Reference

• PR Remove hardcoded Azure Functions templates, use MCP tool #1949: Remove hardcoded Azure Functions templates, use MCP tool #1949
• Functions composition.md (post-MCP): plugin/skills/azure-prepare/references/services/functions/templates/recipes/composition.md
• Parent equity issue: Pass Equity: Azure Compute Skills Lifecycle Parity — 11 PRs open for SME review #1608
• Source PR for Container Apps templates: Gap-2: Container Apps Develop (C → A) — Templates + Dapr + Jobs + Functions-on-ACA #1636

[github-actions]

Thanks for opening this well-scoped enhancement, @paulyuk! 🙌

This issue has been labeled enhancement since it's a clear improvement request, and azure-prepare because the work targets the Container Apps templates within that skill. The assign-to-copilot label has also been applied — a coding agent will look into this and can help with the refactoring effort (deleting hardcoded template files, wiring up the MCP tool calls, and updating the skill to mirror the Functions pattern from PR #1949).

Generated by Issue Triage for issue #2149 · ● 131.6K · ◷

[paulyuk]

@simonjj — co-DRI for this enhancement alongside @paulyuk. You've been the deepest Container Apps contributor on the pass-equity work (PRs #2024, #2025, #2026 fixed up #1636/#1637/#1642), so you're the natural co-owner for the MCP-tool template refactor.

(Couldn't add as a formal assignee — you're an external contributor without collaborator status on this repo. If you want formal assignee status, an org admin can add you. In the meantime, this comment makes the co-ownership clear.)