@katatan/agents
v0.1.12
Published
KATATAN E2E testing agent set — Playwright + KATATAN MCP integration for Claude Code, VS Code, and Cursor
Maintainers
Readme
@katatan/agents
KATATAN's AI agent set.
A single command npx @katatan/agents init sets up a workflow where AI coding agents automatically plan, implement, run, and fix E2E tests.
Overview
What are KATATAN AI Agents?
KATATAN AI Agents are a set of agents for delegating Playwright-based E2E testing to AI.
Just tell the agent "create tests for the login feature" and it will generate, run, and fix test code while operating the browser.
Internally, 5 specialized agents work together:
| Agent | Role |
|---|---|
| katatan-supervisor | The entry point that interacts directly with the user. Interprets instructions and calls other agents |
| katatan-e2e-planner | Explores the browser to investigate the UI. Registers Dictionary entries and TestCases in KATATAN |
| katatan-e2e-generator | Generates Playwright TypeScript test code from TestCases |
| katatan-e2e-runner | Runs tests and records results in KATATAN |
| katatan-e2e-healer | Analyzes failed tests, fixes the code, and re-runs |
What is MCP (Model Context Protocol)?
MCP is a standard protocol for AI agents to connect to external tools and services.
This package configures the following 2 MCP servers:
| Server | Role |
|---|---|
| katatan-test | Manages TestSpecs, TestCases, Dictionaries, and TestReports in KATATAN |
| playwright | Handles browser operations (page navigation, snapshot capture, clicking, etc.) |
Supported AI Clients
| Client | Config File |
|---|---|
| Claude Code | .claude/agents/ + .mcp.json |
| VS Code (GitHub Copilot) | .vscode/mcp.json |
| Cursor | .cursor/mcp.json |
| Kiro (Amazon) | .kiro/settings/mcp.json |
| Google Antigravity | ~/.gemini/antigravity/mcp_config.json (global) + .agents/skills/ |
Setup
Option A: Claude Code Plugin Marketplace (Claude Code only)
The easiest way if you use Claude Code. No npx command required.
1. Set your KATATAN API token as an environment variable
Add the following to your ~/.bashrc, ~/.zshrc, or equivalent:
export KATATAN_API_KEY=kat_xxxxxxxxxxxxxxxxTokens are available from the KATATAN settings page.
2. Add the marketplace and install the plugin
Run the following inside Claude Code:
/plugin marketplace add katatan-org/katatan-TEST
/plugin install katatan-agents@katatan3. Use it
/katatan-agents:katatan-supervisor Create E2E tests for the login pageOption B: npx init (all clients)
Supports Claude Code, VS Code (GitHub Copilot), Cursor, Kiro, and Google Antigravity.
1. Run the setup command
Run the following in the root directory of the project you want to test:
npx @katatan/agents initRequires Node.js 18 or later.
2. Select an AI client
Step 1: Select an AI client
1. Claude Code .claude/agents/ + .mcp.json
2. VS Code (GitHub Copilot) .vscode/mcp.json
3. Cursor .cursor/mcp.json
4. Kiro (Amazon) .kiro/settings/mcp.json
5. Google Antigravity ~/.gemini/antigravity/mcp_config.json + .agents/skills/
Select client(s) (comma-separated, Enter for all) > 1,2You can select multiple clients with comma separation. Pressing Enter alone selects all clients.
3. Enter your KATATAN access token
Step 2: Enter your KATATAN access token
Available from the KATATAN settings page (https://app.katatan.com/)
Token > kat_xxxxxxxxxxxxxxxxAccess tokens can be obtained from the KATATAN settings page.
4. Done
Config files for the selected clients are auto-generated.
If Playwright is not yet installed, you will be prompted to install it.
Generated Files
Claude Code
.claude/
└── agents/
├── katatan-supervisor.md
├── katatan-e2e-planner.md
├── katatan-e2e-generator.md
├── katatan-e2e-runner.md
└── katatan-e2e-healer.md
.mcp.jsonVS Code (GitHub Copilot)
.vscode/
└── mcp.jsonCursor
.cursor/
└── mcp.jsonKiro (Amazon)
.kiro/
└── settings/
└── mcp.jsonGoogle Antigravity
~/.gemini/antigravity/mcp_config.json ← global config (merged with existing)
.agents/
└── skills/
├── katatan-supervisor/SKILL.md
├── katatan-e2e-planner/SKILL.md
├── katatan-e2e-generator/SKILL.md
├── katatan-e2e-runner/SKILL.md
└── katatan-e2e-healer/SKILL.mdUsage
Claude Code
- Open the project in Claude Code
- Talk to
@katatan-supervisor
@katatan-supervisor Create E2E tests for the login pageVS Code (GitHub Copilot)
- Open "MCP: List Servers" in the command palette and enable
katatan-testandplaywright - Give instructions in Copilot Chat (agent mode)
Cursor
- Restart Cursor to apply settings
- Give instructions in Agent mode
Kiro (Amazon)
- Restart Kiro to apply settings
- Give instructions in Agent mode
Google Antigravity
- Restart Antigravity to apply settings
- Talk to the
katatan-supervisorskill
How It Works
User
│ "Create E2E tests for the login feature"
▼
katatan-supervisor (interprets instructions and calls sub-agents)
│
├─▶ katatan-e2e-planner
│ Explore browser → Register Dictionary → Register TestCases
│
├─▶ katatan-e2e-generator
│ TestCases → Generate Playwright TypeScript code → Create TestReport
│
├─▶ katatan-e2e-runner
│ Run tests → Record results in KATATAN
│
└─▶ katatan-e2e-healer (only on failure)
Analyze failure → Fix code → Re-runRequirements
- Node.js 18 or later
- KATATAN account and access token
- One of the supported AI clients
