Home

Claude Code Agent Monitor Handbook

This Wiki is the practical operator handbook for Claude Code Agent Monitor (CCAM). Use it to understand what the dashboard is telling you, investigate agent activity, run a reliable shared deployment, and automate common operations.

This handbook is English-first. For the localized product and architecture tour in English, Vietnamese, Chinese, Korean, and Spanish, open the CCAM static Wiki.

Pick the fastest route

I want to…	Start here
Install CCAM and see my first session	Start Here
Understand the dashboard	Reading the Dashboard
Investigate a session, agent, or tool call	Investigating Sessions and Agents
Operate CCAM for a team	Deployment Decision Guide
Monitor more than one machine	Remote Machines and Data Sources
Automate with the CLI or MCP	CLI Automation Patterns or MCP Workflows
Diagnose a problem now	Troubleshooting Decision Tree

Three ways to use this handbook

Individual user

Follow Choose Your Path, then learn the dashboard, session investigation, task progress, costs, and personal configuration. The recipes are deliberately short enough to use during a live coding session.

Team administrator

Start with the Deployment Decision Guide, then cover Hook Health and Ingestion, Security and Remote Access, backups, observability, upgrades, and incident response.

Power user and integrator

Begin with CLI Automation Patterns, MCP Workflows, and Integration Patterns. Mutating and destructive capabilities are intentionally gated; each guide explains where the safety boundary sits.

What belongs here

The Wiki focuses on decisions, operational habits, investigation techniques, and task-oriented recipes. Exact API contracts and implementation details stay in the versioned repository documentation:

• Documentation index
• Architecture
• API reference
• CLI reference
• Hook reference
• MCP reference

See Documentation Map for the boundary between guides and reference material.

Operating principle

CCAM is local-first. Hook delivery is fail-safe and non-blocking, SQLite remains the system of record, and remote/destructive actions require explicit configuration. When something looks wrong, preserve evidence first, establish whether ingestion or display is stale, and only then change state.

New here? Continue to Start Here →