Karajan Code
Orchestrate multiple AI agents through one automated pipeline. Code, review, test, secure — in a single command.
$
npm install -g karajan-code
click to copy
Orchestrate multiple AI agents through one automated pipeline. Code, review, test, secure — in a single command.
11 roles chained with quality gates. The loop runs until code is approved or the iteration limit is reached.
Chain coder, reviewer, tester, security auditor and more. Each role has its own AI agent and instructions. The orchestrator manages the loop.
When source files change, test changes are required. The coder writes tests first — or the pipeline rejects the iteration.
Static analysis with quality gates. Blocker and critical issues fail the build. Runs locally via Docker — no external services needed.
11 tools exposed via MCP. Use kj from Claude, Codex, or any MCP-compatible host without leaving your agent.
Standard, strict, relaxed, or paranoid. Configure how tough the reviewer should be. Solomon role resolves deadlocks.
Pause and resume sessions. Fail-fast detection stops wasted iterations. Budget tracking monitors tokens and costs per session.
Mix and match. Use Claude as coder, Codex as reviewer — or any combination you prefer.
That's all it takes. The pipeline does the rest.
Use kj as a tool inside your AI agent. Send tasks, get real-time progress, receive structured results.
After npm install -g karajan-code, the MCP server auto-registers in Claude and Codex. Or add it manually:
{
"mcpServers": {
"karajan-mcp": {
"command": "karajan-mcp"
}
}
}