How syamsuddin/odin works
ODIN connects Claude Code on a laptop to Linux servers over SSH stdio MCP. The main workflow is natural-language intent → guarded tool call → project-specific server execution → results returned to Claude for analysis and iteration. The CLI separately manages server/project setup and administration. README capabilities are represented as server-agent responsibilities; the sampled excerpts do not expose every tool handler or its internal wiring.
Main entry points: odin_agent.py MCP Agent · odin_guard.py Risk Guard · odin_mcp_launch.py MCP Launcher · odin-dispatch.sh Forced Dispatcher
Components 20
Laptop Integration
- ODIN CLI
client/odin_cli.py - Risk Guard
client/odin_guard.py - MCP Launcher
client/odin_mcp_launch.py
SSH Transport
- Forced Dispatcher
server/odin-dispatch.sh - Project Launcher
server/run.sh
Server Agent
- MCP Agent
server/odin_agent.py - Linux Commands
server/odin_agent.py - Output Analysis
server/odin_agent.py - Server Inspection
server/odin_agent.py - Deploy Preflight
server/odin_agent.py - Runbook Engine
server/odin_agent.py - Rollback Tracking
server/odin_agent.py
Server State
- Persistent Memory
server/odin_agent.py - Event Journal
server/odin_agent.py - Audit Log
server/odin_agent.py - Watchdog
server/odin_agent.py
Other components
- User
- Claude Code
- SSH
- GitHub Updates
Connections 22
- User → Claude Code: gives intent
- Claude Code → Risk Guard: requests tool
- Risk Guard → Claude Code: allows or asks
- Claude Code → MCP Launcher: spawns MCP
- MCP Launcher → SSH: executes SSH
- ODIN CLI → SSH: provisions server
- SSH → Forced Dispatcher: starts forced command
- Forced Dispatcher → Project Launcher: dispatches project
- Project Launcher → MCP Agent: starts MCP agent
- MCP Agent → Linux Commands: runs commands
- Linux Commands → MCP Agent: returns output
- MCP Agent → Output Analysis: analyzes results
- MCP Agent → Server Inspection: provides inspection
- MCP Agent → Deploy Preflight: checks prerequisites
- MCP Agent → Runbook Engine: serves runbooks
- MCP Agent → Rollback Tracking: tracks state
- MCP Agent → Persistent Memory: reads and writes
- MCP Agent → Event Journal: records events
- MCP Agent → Audit Log: records activity
- MCP Agent → Watchdog: uses watchdog
- MCP Agent → Claude Code: returns MCP results
- Risk Guard → GitHub Updates: checks updates