🏗️ Architecture: How to deploy claude-agent-mcp for server-side code generati...

Date: 2026-01-25
Time: 20:53:28 UTC
Session: editor2-71.75
Agent: backend
Confidence: 90%
Status: ✅ Documented

📋 Decision Context

Question:

How to deploy claude-agent-mcp for server-side code generation on Railway?

Task: Not specified

🎯 Decision Made

Chosen:

Include claude-agent-mcp as subprocess in backend-ai-api container, install Claude Code CLI globally in Dockerfile, spawn via MCPBridgeService STDIO protocol

Options Considered:

• No alternatives documented

💭 Reasoning

Evaluated 3 options: (A) Include in backend-ai-api Dockerfile - simplest, single deployment, works with existing infrastructure; (B) Separate Railway service - more complex, requires HTTP instead of STDIO; (C) BullMQ worker architecture - most scalable but premature. Chose A because: 1) Immediate need is to prove the flow works; 2) Single container reduces operational complexity; 3) STDIO protocol already implemented in MCPBridgeService; 4) Can refactor to workers later when scale is needed. Memory requirement: 2GB min, 4GB recommended.

📁 Files Involved

• deployment/Dockerfile.railway.backend-ai
• deployment/deploy-backend-ai.sh
• apis/backend-ai-api/src/app/workflow/services/mcp-bridge.service.ts

🔗 Related

• Decision ID: dev_1769374408427_qbsh6k8
• Session: 71
• Workspace: editor2

Auto-generated by MCP Smart Consolidation