Getting Started

AgentiHooks is a lifecycle hook system and MCP tool server for Claude Code agents.

What you get

Cost management – slash your token burn with output filtering, file read dedup, lazy MCP loading, context warnings, and native rate limit display. All on by default. See what you save ->
10 lifecycle hooks – intercept every Claude Code event (SessionStart, PreToolUse, Stop, etc.) to log transcripts, inject context, scan for secrets, and learn from tool errors
26 MCP tools across 7 categories – AWS, email, storage, database, compute, observability, and utilities
Profile system – swap agent personalities, skills, and tool access by choosing a profile at install time
3-layer asset merge – agentihooks built-in, bundle global, and profile-specific skills/agents/commands/rules
Sync daemon – auto-propagates new skills, agents, commands, and rules within 60s
Cross-session memory – tool error patterns persist across sessions via NDJSON store
Broadcast system – send operator messages to all active Claude Code sessions simultaneously; severities: info (once, 4h), alert (every turn, 1h), critical (every turn + every tool call, 30m); compose manually or use agentihooks broadcast emit "natural language" for AI-assisted severity selection

Page	What it covers
Installation	Install agentihooks globally into `~/.claude`
Profiles	Choose and switch agent profiles, 3-layer merge
Per-Project Configuration	`.agentihooks.json`, local profiles, MCP whitelists
Portability & Reusability	Move setups across machines, manage credentials