Incident Agent Claude Code

A small TypeScript project for learning and experimenting with Claude Code and AI agent concepts through practical examples.

What This Project Demonstrates

• Claude Code agent workflow
• Agent loops and tool calling
• CLAUDE.md project instructions
• Claude Code permissions and settings
• Project, user, and local settings scopes
• Rules
• Skills
• Subagents
• Git worktrees
• Agent Teams concepts
• MCP server and custom tools

Incident Desk

The demo application manages simple incidents with:

• ID
• Title
• Severity: LOW, MEDIUM, HIGH
• Status: OPEN, INVESTIGATING, CLOSED

It supports operations such as:

• List incidents
• Get incident by ID
• Create incident
• Update incident status
• Delete incident
• Filter high-severity incidents
• Filter open incidents

Claude Code Structure

.claude/
├── agents/
│   └── security-reviewer.md
├── rules/
│   └── tests.md
├── skills/
│   └── add-feature/
│       └── SKILL.md
└── settings.json

CLAUDE.md

CLAUDE.md contains general project instructions.

.claude/rules/ contains instructions for specific parts of the project.

.claude/skills/ contains reusable workflows.

.claude/agents/ contains specialized subagents.

.claude/settings.json contains shared Claude Code settings and permissions.

settings.local.json is ignored because it is user-specific.

MCP Demo

The project includes a small MCP server:

src/mcp.ts

It exposes the Incident Desk functionality to AI applications through MCP.

Example tool:

list-incidents

Test the MCP server:

npx @modelcontextprotocol/inspector npx tsx src/mcp.ts

Connect it to Claude Code:

claude mcp add incident-desk -- npx tsx src/mcp.ts

Then inside Claude Code:

Use the incident-desk MCP tool to list all incidents.

Flow:

Claude Code
    ↓
MCP
    ↓
Incident MCP Server
    ↓
list-incidents tool
    ↓
Incident application

Run Tests

npx vitest run

Main Learning Idea

The project helped connect the main AI-agent concepts:

LLM → reasoning

Tools → actions

Agent loop → reason → act → observe → repeat

Harness → manages the LLM, tools, context and permissions

Agent → the complete working system

MCP → standard way to connect AI applications with external tools and data