MCP Future Work & TODOs

Documentation of remaining work items identified during Session 77 audit.

Session 77 TODO Audit

The following items were identified as future work during the Session 77 code audit and testing verification.

Priority: HIGH

All HIGH priority items resolved in Session 77.

Priority: MEDIUM

1. Vector Similarity Search (Phase 6)

• Location: apis/backend-ai-api/src/app/services/mcp-tools-bridge.service.ts:325
• Description: Implement vector similarity for semantic search in find_similar tool. Currently uses basic string matching; should use embeddings for better semantic matching.
• Effort: 2-3 days
• Dependencies: MongoDB Atlas Vector Search or Pinecone integration
• Benefit: More accurate similarity matching, especially for conceptual queries

2. Conventions Per Space

• Location: apis/backend-ai-api/src/app/services/mcp-tools-bridge.service.ts:367
• Description: Implement space-scoped conventions instead of global conventions. Each space (workspace) should have its own convention set that can inherit from global.
• Effort: 1-2 days
• Dependencies: Space context in MCP session
• Benefit: Allows different coding standards per project/workspace

Priority: LOW

3. Stripe Billing Integration

• Location: apis/backend-ai-api/src/app/services/mcp-billing.service.ts:57, 89, 178-183
• Description: Phase 4 Stripe invoice items - create line items for MCP tool usage, track costs per agent, implement usage-based billing.
• Effort: 3-5 days
• Dependencies: Stripe API keys in production
• Status: Billing service exists, but invoice item creation is stubbed

4. Super Intelligence Patterns

• Location: apis/backend-ai-api/src/app/infrastructure/super-intelligence-gateway.controller.ts:153, 366, 411
• Description:
  • Line 153: Pattern loading from MongoDB for enhanced decision making
  • Line 366: Gap detection in learning pipeline
  • Line 411: Episode counting for reinforcement learning
• Effort: 2-3 days
• Dependencies: Pattern mining data from production usage
• Benefit: Improved AI decision quality over time

5. MCP Tool Definitions Import

• Location: apis/backend-ai-api/src/app/infrastructure/mcp-transport.service.ts:48
• Description: Import actual tool definitions from mcp-gograb-ai package instead of duplicating them. Currently has inline definitions that may drift.
• Effort: 1 day
• Dependencies: mcp-gograb-ai package published to npm or proper workspace linking
• Benefit: Single source of truth for tool schemas

6. Browser Tool Page Object Support

• Location: MCP/mcp-gograb-ai/src/tools/browser.ts
• Description: Add page object pattern support for browser automation. Allow defining reusable page objects for common UI patterns.
• Effort: 2-3 days
• Dependencies: Page object schema definition
• Benefit: More maintainable browser automation tests

7. Task Runner Cron Support

• Location: MCP/mcp-gograb-ai/src/tools/tasks.ts
• Description: Add cron-based scheduling for recurring tasks. Currently only supports immediate and one-time scheduled execution.
• Effort: 1-2 days
• Dependencies: node-cron or similar library
• Benefit: Automated recurring workflows

Resolved in Session 77

• ✅ JWT Token Verification (was HIGH priority)

  • Fixed in apis/backend-ai-api/src/app/mcp-gateway/mcp-gateway.gateway.ts
  • Added proper JWT validation with fallback to configService
  • Imported JwtService and ConfigService

• ✅ Dead Handler Registration (track_outcome, report_module_outcome)

  • Fixed in MCP/mcp-gograb-ai/src/server.ts:2145-2192
  • Added tool definitions and handler switch cases
  • Tools now properly registered and callable

• ✅ Rate Limit Memory Leak (fixed in Session 76)

  • Map size bounded, cleanup on interval
  • No longer accumulates entries unboundedly

• ✅ Index.ts Exports Verified Clean

  • All exports from MCP/mcp-gograb-ai/src/tools/index.ts are used in server.ts
  • No orphaned exports found

Testing Status

Documentation Updates Needed

1. MCP Tool Reference - Add track_outcome and report_module_outcome to tool reference docs
2. Multi-Agent Guide - Document handoff lifecycle and RPC patterns
3. Error Handling Guide - Document expected error responses and retry strategies

Architecture Considerations

Session 78+ Recommendations

1. Tool Count Monitoring: Currently at 78 tools. MCP clients have soft limits (~80-100 tools). Consider consolidation if adding more.

2. Backend Health Checks: Add periodic health check from MCP to backend. Currently fails silently on backend unavailability.

3. Metrics Dashboard: Create observability dashboard for:

   • Tool usage frequency
   • Error rates by tool
   • Response latencies
   • Agent online/offline events

4. Rate Limiting Per Agent: Current rate limiting is global. Consider per-agent limits for better isolation.

Last updated: Session 77 - January 31, 2026 Author: AI Code Review (Session 77 Phase 4)