// HOW IT WORKS

Three steps to
programmable browsing.

Compile the runtime, expose CDP on localhost, and connect with the Koko SDK or any CDP client. No Chromium monolith required for everyday automation.

// HOW IT WORKS

Three steps to
programmable browsing.

quickstart.ts
1git clone https://github"text-primary">.com/ivanUri/koko-core
2cd koko-core
3"text-foreground">zig build
Ready

// PROTOCOLS & TOOLING

Connect agents,
scripts, and infra.

Koko speaks the protocols automation systems already use — CDP for scripts, MCP for agents, and a first-party SDK for programmatic control.

  ┌─┐
  │C│
  └─┘

CDP

Protocol

→
  ╔═╗
  ║M║
  ╚═╝

MCP

Agents

→
  ┌▶┐
  └─┘

Koko SDK

Client

→
  [TS]
  [TS]

TypeScript

Language

→
  ◈◈
  ◈◈

Zig

Engine

→
  ≋≋
  ≋≋

V8

JS runtime

→
  {f}
  ---

koko-fetch

CLI

→
  ▲
  ─

Node.js

Tooling

→

Agent-oriented extraction

Token-efficient page state for LLM workflows — markdown, semantic trees, structured data, and stable backendNodeId handles.

SDK documentation
// Agent extraction
const md = await page.markdown();
const tree = await page.semanticTree({
format: "text",
maxDepth: 4
});

Ready to write your first script?

Install koko-sdk, connect to a running CDP server, and automate pages with Playwright-style APIs plus agent extraction helpers.

Explore the SDKQuickstart guide