DISCOVER THE FUTURE OF AI AGENTS

OpenCLI

Added May 4, 2026
Agent & Tooling
Open Source
TypeScriptNode.jsElectronModel Context ProtocolAI AgentsBrowser AutomationCLIAgent & ToolingDeveloper Tools & CodingAutomation, Workflow & RPAProtocol, API & Integration

An AI-native CLI Hub and runtime that transforms websites, browser sessions, Electron apps, and local tools into deterministic CLIs.

OpenCLI is a unified command-line hub designed for both humans and AI Agents. Its core mechanism uses a local Daemon paired with a Chrome extension to reuse authenticated browser sessions, converting website operations into deterministic CLI commands with zero LLM consumption. The project includes 100+ built-in site adapters (e.g., Bilibili, Reddit, HackerNews, Twitter/X, Amazon, Gemini, Claude) supporting article scraping and Markdown conversion. For AI coding assistants like Claude Code and Cursor, OpenCLI provides low-level browser automation primitives (opencli browser) and standardized SKILL.md descriptions, enabling Agents to perform complex web interactions like clicking, typing, and data extraction. Furthermore, it supports direct control of Electron desktop apps like Cursor, ChatGPT, and Notion via the CDP protocol, and acts as a unified entry point to aggregate and manage fragmented local CLI tools (gh, docker, obsidian, etc.). The system is built with TypeScript (ESM), runs on Node.js ≥ 21, and features a comprehensive plugin system, multi-browser Profile bridging, and automated adapter development workflows.

Core Capabilities

  • Site Adapters & Content Extraction: 100+ built-in adapters, deterministic execution, zero token consumption, pipeline and CI friendly
  • Browser Automation Primitives: open, state, click, type, select, wait, extract, network and more
  • Desktop App Control: Drive Electron applications via CDP protocol
  • Unified CLI Hub: Discover, auto-install, and proxy-execute external CLIs
  • Security: Credentials never leave the browser process

Architecture Highlights

  • Communication layer: Local Daemon (HTTP port 19825) + Chrome extension + WebSocket + CDP four-layer bridging
  • Adapter layer: YAML definition + TypeScript implementation, unified registration via cli-manifest.json
  • Agent Skill system: SKILL.md standardized descriptions loaded by AI coding assistants
  • Module exports: registry-api, browser/cdp, browser/page, download, pipeline sub-modules
  • Build & test: TypeScript (tsc) compilation, vitest covering unit/extension/adapter/e2e test projects

Related Projects

View All

STAY UPDATED

Get the latest AI tools and trends delivered straight to your inbox. No spam, just intelligence.