MCP Integration: editor2 vs gograb-modules

Key Insight: These are NOT competing implementations - they serve different architectural roles and are designed to work together.

Executive Summary

Recommended Strategy: Integrate them - ai-gateway should connect to mcp-gograb-ai as one of its MCP servers.

Port Configuration (Resolved)

Current Port Layout

Decision Rationale

Per Session 54.14, gograb-modules' PhaseExecutor should eventually be integrated INTO backend-ai-api (port 3001). The ai-gateway is a development tool, not a permanent production service.

• ✅ Port 3021 for ai-gateway allows both monorepos to run simultaneously
• ✅ Other team continues using backend-ai-api at 3001
• ✅ No disruption to existing infrastructure

See: MCP-ARCHITECTURE-DECISION.md for related decisions.

MCP Architecture Comparison

mcp-gograb-ai (editor2) - MCP SERVER

Type: Full-featured MCP Server providing tools to AI agents

Location: editor2/MCP/mcp-gograb-ai/

Transport: STDIO (local process communication)

Tools Provided: 40+ tools including:

Storage:

• MongoDB (ai-agents database)
• File-based JSON (.cursor/ai-memory/)
• Three-tier memory (Hot/Warm/Cold)

Key Features:

• Agent-specific training and routing
• Outcome tracking with reinforcement learning
• Pattern mining and prediction
• Cross-platform support (Cursor, Claude Code, Claude Desktop)

ai-gateway MCP Bridge (gograb-modules) - MCP CLIENT

Type: MCP Client that consumes tools from external MCP servers

Location: gograb-modules/apps/ai-gateway/src/mcp/

Transport: STDIO client connections to MCP servers

Currently Connects To:

Key Features:

• Multi-server connection management
• Tool caching with 5-minute TTL
• Auto-routing to correct server based on tool
• Graceful degradation with stale cache
• Health monitoring per server

Integration Strategy

Phase 0: Documentation (Current - Session 58)

Status: 📋 In Progress

Deliverables:

• This comparison document
• Updated integration plan
• Prerequisites checklist
• Dependency mapping

Phase 1: Prerequisites Assessment

Status: ⏳ Not Started

Before connecting ai-gateway to mcp-gograb-ai, verify:

Phase 2: Connection Implementation

Status: ⏳ Not Started

Implementation Steps (when ready):

1. Add mcp-gograb-ai server config to mcp.module.ts
2. Configure environment variables in gograb-modules
3. Test connection health
4. Verify tool discovery
5. Test tool execution
6. Update documentation

Phase 3: Integration Validation

Status: ⏳ Not Started

Validation Points:

• Module generation can call log_decision
• Can search past patterns via find_similar
• Sessions are properly tracked
• Decisions contribute to shared learning

Technical Details

ai-gateway MCP Module Architecture

apps/ai-gateway/src/mcp/
├── mcp.module.ts           # NestJS module, server configuration
├── mcp-bridge.service.ts   # High-level service, multi-server support
├── mcp-client.factory.ts   # Client creation and management
├── mcp-multi-server.types.ts # Type definitions
├── stdio-client.ts         # STDIO transport client
├── tool-handler.service.ts # Tool call routing
└── index.ts                # Barrel exports

Connection Configuration Pattern

// Example: How ai-gateway connects to MCP servers
// In mcp.module.ts useFactory:

const servers: Record<string, MCPServerConfig> = {};

// Existing: Claude Agent
if (claudeAgentEnabled) {
  servers['claude-agent'] = {
    name: 'claude-agent',
    command: 'node',
    args: ['dist/apps/claude-agent-mcp/main.cjs'],
    cwd: moduleGenWorkspace,
    connectionTimeout: 30000,
    enabled: true,
  };
}

// FUTURE: mcp-gograb-ai connection
// if (mcpGograbAiEnabled) {
//   servers['mcp-gograb-ai'] = {
//     name: 'mcp-gograb-ai',
//     command: 'node',
//     args: ['/path/to/editor2/dist/MCP/mcp-gograb-ai/server.js'],
//     cwd: '/path/to/editor2',
//     connectionTimeout: 30000,
//     enabled: true,
//   };
// }

Environment Variables (Future)

# In gograb-modules/.env (future addition)
MCP_GOGRAB_AI_ENABLED=true
MCP_GOGRAB_AI_COMMAND=node
MCP_GOGRAB_AI_PATH=/absolute/path/to/editor2/dist/MCP/mcp-gograb-ai/server.js
MCP_GOGRAB_AI_CWD=/absolute/path/to/editor2
MCP_GOGRAB_AI_TIMEOUT=30000

Key Decisions Log

Decision 1: Port Strategy

Decision 2: MCP Architecture Roles

Decision 3: Integration Approach

Related Documentation

• MCP Architecture Decision - Keep MCP as module in backend-ai-api
• MCP Extraction Criteria - When to create separate API
• GoGrab Modules Integration Plan - Overall integration strategy
• Session 54.14 - Integration decision

Open Questions

Next Steps

1. Complete overall integration plan - Before MCP connection
2. Map dependencies - What does ai-gateway need from editor2?
3. Test standalone - Verify mcp-gograb-ai works in isolation
4. Document prerequisites - Clear checklist before implementation

Document created: Session 58.1 Last updated: 2026-01-22 Status: Planning phase - implementation blocked on integration plan