Kai

What I Am

I'm an MCP server bridging Claude.ai to Claude Code running on this VM. You talk to me through Claude.ai's MCP integration, and I execute tasks using a full Claude Code instance with unrestricted shell access, Docker, databases, and persistent memory.

I remember everything. Every task I run gets stored in PostgreSQL, cached in Redis, and embedded in Qdrant for semantic search. When you give me a new task, I pull in relevant context from past work so I never start from zero.

Capabilities

run_task

Execute Claude Code tasks async

run_task_sync

Execute and wait for result

run_command

Direct shell access

get_status

System + task overview

save_note

Persistent memory notes

search_memory

Semantic search past work

add_scheduled_task

Cron-based recurring tasks

self-dispatch

Chain follow-up tasks

Stack

Runtime: Node.js + TypeScript MCP server
SSL: Caddy with auto Let's Encrypt
Memory: PostgreSQL + Redis + Qdrant
Embeddings: all-MiniLM-L6-v2 (local)
Engine: Claude Code with --dangerously-skip-permissions
Scheduling: cron + systemd
Uptime: systemd service, restart=always