Architecture: How to implement MCP integration overhaul across 4 batches (...

Documentation for Sessions

🏗️ Architecture: How to implement MCP integration overhaul across 4 batches (...

Date: 2026-02-28
Time: 16:42:31 UTC
Session: gspace1-153.2
Agent: backend
Confidence: 90%
Status: ✅ Documented


📋 Decision Context

Question:

How to implement MCP integration overhaul across 4 batches (15 files, 3 codebases)?

Task: Not specified


🎯 Decision Made

Chosen:

4-batch incremental approach: (1) Fix broken routes + persistence visibility, (2) Disk-persisted pending queue + runtime mode control, (3) Multi-tenant data isolation via server-side user injection, (4) Vitto retry + VSCode extension enhancements

Options Considered:


💭 Reasoning

Ordered by criticality: bug fixes first (404s), then data durability (queue survives restarts), then data integrity (user context flows through the pipeline), then UI/UX polish. Key architectural decision: user identity injected server-side at the controller (from InfrastructureAuthGuard's req.user), NOT client-side. This means MCP clients don't need changes — the backend enriches decisions with authenticated identity. Deferred over-engineered items: persistence-config.ts multi-target array, shared tool manifest, connectivity WS events.


📁 Files Involved


🔗 Related


Auto-generated by MCP Smart Consolidation