Multi-Agent AI Infrastructure · 3 Machines · 5 CLI Agents · 67 Docker Containers · 140+ Repos
| From | Alias | Target | User | Port |
|---|---|---|---|---|
| WSL (rod-ml) | ssh rod-server | 10.0.1.6 | rod | 22 |
| WSL (rod-ml) | ssh mac-mini | 10.0.1.35 | rodchemist | 22 |
| WSL (rod-ml) | ssh rod-5050 | 10.0.1.11 | sshrod | 22 |
| rod-server | ssh rod-ml | 10.0.1.3 | rod | 2223 |
| rod-server | ssh mac-mini | 10.0.1.35 | rodchemist | 22 |
| Mac mini | ssh rod-server | 10.0.1.6 | rod | 22 |
| Mac mini | ssh rod-ml | 10.0.1.3 | rod | 2223 |
All agents share the same workspace at /mnt/c/rod_hub/ on rod-ml, with agent-specific instruction files and a universal AGENTS.md.
Anthropic's CLI agent. Primary agent with full MCP ecosystem, 55+ cloud tools, and deep infrastructure control.
Google's CLI agent. Strong long-context capabilities, good for large codebases and cross-file analysis.
Alibaba's CLI agent. Similar toolset to Claude with bash, read, write, glob, grep, and task tools.
OpenAI's sandboxed CLI agent. Features apply_patch, spawn_agent, and multi_tool_use.parallel capabilities.
Go-based TUI agent. Supports interactive, non-interactive (opencode run), and headless server modes.
┌────────────────────────────────────────────────────────┐
│ AGENTS.md (Universal) ← All agents read this first │
├────────────────────────────────────────────────────────┤
│ ├── CLAUDE.md (v3.17) ← Full infra + MCP transport │
│ ├── GEMINI.md (v2.1) ← Gemini-specific │
│ ├── QWEN.md (v2.1) ← Qwen-specific │
│ ├── CODEX.md (v1.1) ← Codex-specific │
│ └── OPENCODE.md (v1.1) ← OpenCode-specific │
├────────────────────────────────────────────────────────┤
│ ~/.claude/CLAUDE.md (Global) ← Per-machine awareness │
│ ├── rod-ml → Infrastructure + workspace │
│ ├── rod-server → Docker stacks + Traefik API │
│ └── Mac mini → SSH mesh + remote access │
└────────────────────────────────────────────────────────┘
Model Context Protocol servers extend all 5 CLI agents + Claude Desktop with tools for infrastructure, research, automation, and more. All 6 agents share the same core MCP servers.
| Server | Transport | Status | Purpose |
|---|---|---|---|
| Filesystem | stdio (npx) | ✓ All agents | File browsing across repos and guides |
| Google Workspace | stdio (uv run) | ✓ All agents | 137 tools: Gmail, Drive, Calendar, Docs, Sheets, Tasks, Contacts |
| Web Scraper | stdio (python3) | ✓ All agents | Crawl4AI markdown, raw HTML+CSS selectors, Playwright screenshots |
| n8n | HTTP/SSE | ✓ All agents | n8n workflow automation (7,000+ integrations) |
| Infra MCP | stdio (node) | ✓ All agents | Docker, Traefik, DNS, services (infrastructure queries) |
Self-hosted MCP servers on rod-server (10.0.1.6) via _mcp-tools docker-compose. Each has a .home Traefik route. Verified with real API keys.
| Server | Traefik Route | Transport | Tools | Status |
|---|---|---|---|---|
| Grafana MCP | grafana-mcp.home | SSE (supergateway) | 54 tools — dashboards, alerts, Loki, Prometheus, Pyroscope, oncall | Connected |
| HuggingFace MCP | huggingface-mcp.home | Streamable HTTP | 14 tools — model search, papers, Spaces, docs, image gen | Connected |
| PubMed MCP | pubmed-mcp.home | SSE (supergateway) | 5 tools — article search, fetch, connections, charts, research agent | Connected |
| Mermaid MCP | mermaid-mcp.home | SSE (native) | 1 tool — diagram rendering (PNG/SVG via Playwright) | Connected |
| Server | Endpoint | Transport | Purpose |
|---|---|---|---|
| Socket Security | npx @socketsecurity/mcp | stdio | Dependency security scanning |
| PDF Tools | npx @mcp-apps/pdf-tools-mcp-server | stdio | PDF read, merge, split, watermark |
| Make.com | eu2.make.com/mcp | SSE (HTTPS) | Make.com automation (3,000+ apps) |
| Scholar Gateway | connector.scholargateway.ai/mcp | Streamable HTTP | Academic paper search (Wiley CONNECT) |
| CanLII | python3 mcp/canlii/mcp_canlii.py | stdio | Canadian legal case law search |
| Connector | URL | Status | Tools |
|---|---|---|---|
| Microsoft 365 | microsoft365.mcp.claude.com | Connected | Outlook, SharePoint, OneDrive |
| Crypto.com | mcp.crypto.com | Connected | Market data, tickers, candlesticks |
| Cloudflare | bindings.mcp.cloudflare.com | Connected | Workers, KV, R2, D1, DNS |
| HuggingFace (Cloud) | huggingface.co/mcp | Connected | Models, datasets, papers, spaces |
| Indeed | mcp.indeed.com | Connected | Job search, job details |
| Figma | mcp.figma.com | Needs Auth | Design files, components |
| Sentry | mcp.sentry.dev | Needs Auth | Error tracking, issues |
Each agent has different transport capabilities. Bridge strategy standardized after testing all combinations.
| Agent | Docker LAN MCPs (SSE/HTTP) | Remote HTTPS MCPs |
|---|---|---|
| Claude Code | Native type: "sse" | Native SSE or httpUrl |
| Claude Desktop | Native SSE (via wsl.exe + supergateway) | wsl.exe + mcp-remote |
| Qwen | supergateway --sse (httpUrl only supports Streamable HTTP) | httpUrl for Streamable HTTP |
| Gemini | supergateway --sse / --streamableHttp | mcp-remote for HTTPS |
| Codex | codex mcp add + supergateway (ignores config.toml) | codex mcp add + mcp-remote |
| OpenCode | supergateway --sse / --streamableHttp | mcp-remote for HTTPS |
supergateway is the correct bridge for LAN HTTP/SSE → stdio. Use --sse for SSE, --streamableHttp for HTTP.mcp-remote only works with HTTPS. Blocks non-HTTPS; EventSource fails on LAN SSE servers.Primary host: AMD Ryzen 5 4600H, 62GB RAM, 935GB NVMe, GTX 1650 4GB, Ubuntu 24.04.3 LTS
| Service | Container | Port(s) |
|---|---|---|
| Traefik (Reverse Proxy) | traefik | 80, 443, 8080 |
| Pi-hole DNS | pihole | 53, 8053 |
| Portainer | portainer | 9000 |
| Nextcloud | nextcloud | via Traefik |
| Forgejo (Git) | forgejo | 3000, 2224 |
| n8n Workflows | n8n | 5678 |
| Home Assistant | homeassistant | 8123 |
| Ollama LLM | ollama | 11434 |
| Grafana | grafana | 3000 |
| Prometheus | prometheus | 9090 |
| Frigate NVR | frigate | via Traefik |
| File Browser | filebrowser | via Traefik |
| Kavita (eBooks) | kavita | via Traefik |
| qBittorrent+Gluetun | qbittorrent | via Traefik |
| MariaDB | mariadb | 3306 (internal) |
| Redis | redis | 6379 (internal) |
| MCP Proxy | mcp-proxy | via Traefik |
| Terminal (4x) | terminal-* | via Traefik |
| RustDesk (2x) | rustdesk-* | 21115-21119 |
| Crypto Predictor | crypto-* | via Traefik |
| Service | Container | Port |
|---|---|---|
| API Gateway | librosynth-api-gateway | 8000 |
| Orchestrator | librosynth-orchestrator | — |
| Meta-Learning | librosynth-meta-learning | — |
| Knowledge Synthesis | librosynth-knowledge | — |
| Intelligence Metrics | librosynth-metrics | — |
| Optimization | librosynth-optimization | — |
| Web Dashboard | librosynth-web-dashboard | 3002 |
| PostgreSQL | librosynth-postgres | 5433 |
| RabbitMQ | librosynth-rabbitmq | 5672, 15672 |
| MinIO | librosynth-minio | 9002, 9003 |
| Service | Port |
|---|---|
| Audiobookshelf | 13378 |
| Plex Media Server | 32400 |
Every service has both .home (Local CA) and .sanchezfamily.ca (Let's Encrypt) routes.
| Service | .home URL | .sanchezfamily.ca URL |
|---|---|---|
| Homepage | dashboard.home | dashboard.sanchezfamily.ca |
| Nextcloud | nextcloud.home | nextcloud.sanchezfamily.ca |
| Forgejo (Git) | gitlab.home | gitlab.sanchezfamily.ca |
| n8n Workflows | n8n.home | n8n.sanchezfamily.ca |
| Home Assistant | homeassistant.home | homeassistant.sanchezfamily.ca |
| Ollama LLM | ollama.home | ollama.sanchezfamily.ca |
| Grafana | grafana.home | grafana.sanchezfamily.ca |
| Portainer | portainer.home | portainer.sanchezfamily.ca |
| Pi-hole DNS | pihole.home | pihole.sanchezfamily.ca |
| File Browser | files.home | files.sanchezfamily.ca |
| Kavita | kavita.home | kavita.sanchezfamily.ca |
| Frigate NVR | frigate.home | frigate.sanchezfamily.ca |
| qBittorrent | downloads.home | qbittorrent.sanchezfamily.ca |
| Crypto Predictor | crypto.home | crypto.sanchezfamily.ca |
| Traefik Dashboard | traefik.home | traefik.sanchezfamily.ca |
| LibroSynth Dashboard | librosynth.home | librosynth.sanchezfamily.ca |
| LibroSynth API | librosynth-api.home | librosynth-api.sanchezfamily.ca |
| MCP Proxy | mcp.home | mcp.sanchezfamily.ca |
| Terminal Hub | terminal.home | terminal.sanchezfamily.ca |
| Main Website | sanchez.home | sanchezfamily.ca |
| Grafana MCP | grafana-mcp.home | — |
| HuggingFace MCP | huggingface-mcp.home | — |
| PubMed MCP | pubmed-mcp.home | — |
| Mermaid MCP | mermaid-mcp.home | — |
| Audiobookshelf | audiobooks.home | — |
| Plex | plex.home | — |
| Domain Pattern | Provider | Renewal |
|---|---|---|
*.home | Sanchez Home Network CA (Local) | Manual (trusted on all machines) |
*.sanchezfamily.ca | Let's Encrypt | Auto-renewed by Traefik |
| Database | Internal Port | Host Port | Access |
|---|---|---|---|
| MariaDB | 3306 | — | Internal only |
| PostgreSQL (LibroSynth) | 5432 | 5433 | LAN: 10.0.1.6:5433 |
| Redis | 6379 | — | Internal only |
| RabbitMQ | 5672 / 15672 | 5672 / 15672 | AMQP + Management UI |
| MinIO | 9002 / 9003 | 9002 / 9003 | S3-compatible storage |
High-level overview below. For per-repo architecture docs, agent usage guides, and LLM-readable references, see docs.home (37 documented repos with interactive diagrams).
| Location | Purpose | Count |
|---|---|---|
/mnt/c/rod_hub/repos/ | Active development workspace | ~20 repos |
/mnt/c/rod_hub/services/ | Service implementations (MCP, exporters, tools) | ~10 services |
/mnt/e/github/ | Production repositories | 140+ repos |
/mnt/e/github/Claude_Repos/ | Claude/Anthropic SDK & reference repos | 25 repos |
gitlab.sanchezfamily.ca | Forgejo self-hosted mirrors | Mirrors |
| Prefix | Standard | Coverage | Required Files |
|---|---|---|---|
emp_ | Empiric | 90%+ | 10 files, full audit trail |
rod_, exp_ | RodChemist | 80%+ | 6 files, pragmatic |
ext_, clon_, fork_ | External | — | README + TODO + agent_scripts |
132+ AI agent instances with LangGraph orchestration, iterative learning (325% quality improvement), and cross-LLM knowledge sharing.
Location: /mnt/e/github/Rod-Agents/
129 time series forecasting models across 5 categories: Classical (14), State-Space (21), Graph Neural (16), Foundation (7), Transformers (18).
Location: /mnt/e/github/ai_trainer_v02/
6-agent autonomous competition system with 10-phase lifecycle. Agent pairs: Claude (Strategic + QA), GPT (Code + Model), Qwen (Data + Math).
Location: /mnt/c/_repos_/kaggle_automation/
Self-improving AI platform with evolutionary knowledge synthesis. 10 Docker services, 4 synthesis modes: CROSS_DOMAIN, CAUSAL, EMERGENT, CONTRASTIVE.
Deployed: [email protected]:/home/rod/_rod/_librosynth/
Multi-AI conversation indexing with 220K+ messages and GPU-accelerated vector search across Claude, ChatGPT, Gemini, Qwen, Codex.
Location: /mnt/e/github/Rod-Agents/instances/unified_ai_kb/
137 tools across 12 Google services: Gmail, Drive, Calendar, Docs, Sheets, Tasks, Contacts, and more. OAuth2 authenticated.
Location: /mnt/c/rod_hub/services/google_workspace_mcp/
Collection of 25 Anthropic, OpenAI, and AI reference repositories at E:\github\Claude_Repos\
Anthropic's agentic coding tool (this CLI). Source code for Claude Code.
Origin: github.com/anthropics/claude-code
Official Anthropic Agent SDK for building custom AI agents with tool use and orchestration.
Origin: github.com/anthropics/claude-agent-sdk-python
Demo applications and examples for the Claude Agent SDK.
Origin: github.com/anthropics/claude-agent-sdk-demos
Official Python SDK for the Anthropic API (Claude models, tool use, streaming).
Origin: github.com/anthropics/anthropic-sdk-python
Official Go SDK for the Anthropic API.
Origin: github.com/anthropics/anthropic-sdk-go
Official C# SDK for the Anthropic API.
Origin: github.com/anthropics/anthropic-sdk-csharp
Official PHP SDK for the Anthropic API.
Origin: github.com/anthropics/anthropic-sdk-php
Recipes, guides, and example code for building with Claude (prompting, RAG, agents).
Origin: github.com/anthropics/anthropic-cookbook
Quick-start examples for Claude API integration patterns.
Origin: github.com/anthropics/anthropic-quickstarts
Official Claude Code plugin implementations and extension framework.
Origin: github.com/anthropics/claude-plugins-official
Knowledge work MCP plugins for Claude (document processing, research tools).
Origin: github.com/anthropics/knowledge-work-plugins
Claude Code skill definitions and examples.
Origin: github.com/anthropics/skills
Claude AI MCP server implementations and reference architecture.
Origin: github.com/anthropics/claude-ai-mcp
GitHub Action for running Claude Code in CI/CD pipelines.
Origin: github.com/anthropics/claude-code-action
Base GitHub Action for Claude Code integrations.
Origin: github.com/anthropics/claude-code-base-action
Healthcare AI use cases and reference implementations with Claude.
Origin: github.com/anthropics/healthcare
Life sciences AI applications (genomics, drug discovery) with Claude.
Origin: github.com/anthropics/life-sciences
CLI Proxy API for bridging local commands to cloud services.
Origin: Community
Experimental C compiler written by Claude as a demonstration of code generation capability.
Origin: github.com/anthropics/claudes-c-compiler
Curated collection of awesome LLM application examples and patterns.
Origin: Community collection
Dev container features for Claude Code development environments.
Origin: github.com/anthropics/devcontainer-features
OpenAI Codex CLI source code. Rust-based with TOML config and sandbox modes.
Origin: github.com/openai/codex
Open-source RAG engine with deep document understanding and multi-model support.
Origin: github.com/infiniflow/ragflow
Workflow automation platform (self-hosted). Fair-code licensed, 400+ integrations.
Origin: github.com/n8n-io/n8n
Full Google Workspace integration: Gmail, Drive, Calendar, Docs, Sheets, Tasks, Contacts. 137 tools, OAuth2.
AI conversation exporter supporting ChatGPT, Claude, and Grok formats.
Config persistence system. NTFS-based backup surviving WSL corruption. Real-time watcher + cron + session hooks.
LimeSurvey integration MCP server for survey management.
Google Forms automation and management tools.
OCR automation service for document processing.
┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
│ 1.ANALYZE │───▶│ 2.RETRIEVE│───▶│ 3.REASON │───▶│ 4.EXECUTE │
└──────────┘ └──────────┘ └──────────┘ └──────────┘
│
┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ 7.UPDATE │◀───│ 6.CRITIQUE│◀───│ 5.SYNTH │◀──────────┘
└──────────┘ └──────────┘ └──────────┘
LLM Fallback Chain: Claude → Codex → Gemini → Qwen → vLLM → Ollama
| Mode | Use Case | Example |
|---|---|---|
CROSS_DOMAIN | Combine insights from different fields | Financial + ML analysis |
CAUSAL | Root cause analysis | Why did model X fail? |
EMERGENT | Novel research discovery | Unexpected pattern in data |
CONTRASTIVE | Trade-off analysis | LSTM vs Transformer comparison |
| Directory | Size | Contents |
|---|---|---|
/mnt/f/GDriveFiles/Takeout/DATA/ | 550 GB | Primary data (46GB emails) |
/mnt/f/GDriveFiles/Takeout/Google Photos/ | 100 GB | Photo library |
/mnt/f/GDriveFiles/Takeout/Location History/ | 2 GB | Timeline & location data |
| Total: 652 GB, 450K+ files, 122K+ JSON files, 35+ Google services | ||
Location: /mnt/c/rod_hub/services/claude_central/ — NTFS-based, survives WSL disk corruption.
| Layer | Mechanism | Frequency |
|---|---|---|
| Real-time watcher | inotifywait daemon (watch.sh) | 5-second debounce, always running |
| Periodic cron | crontab | Every 6 hours |
| Session hook | Claude session start | Every new Claude session |
| File | Source | Purpose |
|---|---|---|
| AGENTS.md, CLAUDE.md, GEMINI.md, QWEN.md, CODEX.md, OPENCODE.md | /mnt/c/rod_hub/ | Agent instruction files |
| MCP configs (3 files) | ~/.claude/ | MCP server definitions |
| Credentials, session state | ~/.claude/ | Auth tokens, settings |
claude-sync # Full sync (backup + push + snapshot)
claude-sync --backup # Backup WSL critical files to NTFS
claude-sync --restore # Restore WSL from NTFS backup
claude-sync --quiet # Auto-runs on Claude session start
claude-watch # Start real-time file watcher daemon
claude-watch --stop # Stop watcher
claude-watch --status # Check watcher status
infra-status # Full infrastructure report
infra-status --quick # System + containers only
infra-status --json # Machine-readable JSON