🏗️ Architecture: How should we enhance the MCP Command Center to distinguish ...

📋 Decision Context

Question:

How should we enhance the MCP Command Center to distinguish execution agents vs training agents and support multi-platform, multi-session scenarios?

Task: Supercharge MCP system with agent distinction, real-time memory tiers, multi-session support

🎯 Decision Made

Chosen:

Implement two-layer agent visualization: Execution Layer (Claude/Cursor actively working) vs Training Layer (MCP agents learning). Add real-time memory tier tracking by emitting events from tiered-memory.ts. Support multiple concurrent sessions with array-based task tracking. Already platform-agnostic naming (cursor- prefix only for UI compat).

Options Considered:

• No alternatives documented

💭 Reasoning

After deep analysis, discovered agent naming is ALREADY platform-agnostic in agent-router.ts - only Cursor gets cursor- prefix for UI compatibility. The real issue is UI doesn't distinguish between: 1) Execution agents (Claude/Cursor writing code NOW) vs 2) Training agents (MCP specialized agents learning from decisions). Memory tiers show 0 because no events emitted when getHotDecisions/getWarmDecisions/getColdDecisions called. Multi-session needed because Cursor + Claude Code can run simultaneously. Solution: Two separate UI panels for execution vs training, emit memory tier events, track multiple active tasks by sessionID. This creates clear mental model: AI assistant executes → MCP agents learn → Platform AI improves.

📁 Files Involved

• docs/english/sessions/editor2/session-34/README.md
• sessions/sessions-history/editor2/session-34.json

🔗 Related

• Decision ID: dev_1767982067422_od0ip3a
• Session: 31
• Workspace: editor2

Auto-generated by MCP Smart Consolidation