EI

eze-is/web-access

Developer tools

This page was translated automatically. Technical details may differ from the original. View the original

9.3K stars Quality 63 Trend 58

Equip Claude Code with full networking capabilities: Layer 3 channel scheduling + browser CDP + parallel partition and conquer

Overview

Equip AI Agents with comprehensive networking capabilities. 🌐 Official Website · 📖 Design Explanation · ⚡ Fast Installation The original networking capabilities of AI Agents (WebSearch, WebFetch) lack scheduling strategies and browser automation capabilities. This Agent Skill supplements this: Compatible with all SKILL.md-supporting Agents (Claude Code, Cursor, Gemini CLI, Codex CLI, etc.). Recommended must-read: Web Access: A Skill That Maximizes Agent Networking and Browser Capabilities. This book provides a comprehensive introduction to the development details of Web-Access Skills and the design philosophy of Agent Skills, helping you write skills with similar general and high ceiling requirements - — CDP Proxy no longer binds to Chrome and adds Edge adaptation (Chromium, Chrome Canary, and other Chromium systems are integrated through the same automatic discovery mechanism). In edge://inspect/#remote-debugging, check "Allow remote debugging for this browser instance" - — Add config.env (gitignored, created from template on first run), fix the default browser via WEB_ACCESS_BROWSER; When multiple browsers enable toggle simultaneously, the agent will ask for preferences. Also supports single overwrite --browser - — If a preference/specified browser is not launched or toggle is not enabled, a hard error is provided with clear handling steps and it will not quietly connect to other browsers; After the first successful connection, the proxy pins the browser ID to avoid running drift - — Local bookmark/history search by default traverses Chrome and Edge, using --browser to limit a single browser - — Added scripts/find-url.mjs to locate URLs by keyword/timewindow/visit frequency from local Chrome bookmarks/history.

Overview (machine translated)

🌐 Language

English | Chinese Simplified | Traditional Chinese | Japanese | 한국어 | Français | Deutsch | Español | Português | Русский

Equip AI Agents with fully connected skills.
🌐 Official Website · 📖 Design Explanation · ⚡ Fast installation

The AI Agent’s original networking capabilities (WebSearch, WebFetch) lack scheduling strategies and browser automation capabilities. This Agent Skill supplements: Network Connection Strategy + CDP Browser Operation + Site Experience Accumulation. Compatible with all SKILL.md-supporting Agents (Claude Code, Cursor, Gemini CLI, Codex CLI, etc.).

Recommended must-read: Web Access: A Skill That Maximizes Agent Networking and Browser Capabilities. This book provides a comprehensive introduction to the development details of Web-Access Skills and the design philosophy of Agent Skills, helping you write skills with similar general-purpose capabilities and high ceilings

[! IMPORTANT] Before upgrading from the old version to v2.6, please send the following message to your agent (Browser preferences and site experience will be changed to ~/.web-access/, npx skills update will first delete the old directory; if not moved first, it will be lost):

installed with git or Claude Code plugins, the upgrade will automatically migrate; Repeat this again as safe (copy only, do not overwrite). See references/migration-2.6.md for details.


v2.6.0 Capabilities

Ability Description
Networking tools automatically select WebSearch / WebFetch / curl / Jina / CDP, autonomously judges based on the scenario, allowing arbitrary combinations
CDP Proxy browser operation Directly connects to users’ daily browsers (Chrome / Edge / Brave / Arc / DIA or any Chromium kernel browser), naturally supports login mode, supports dynamic pages, interactive operations, video screenshots
Click and iframe /click(JS click), /clickAt (CDP real mouse event), /clickXY (coordinate click, can access cross-site iframe, disable tabs), /setFiles (file upload); /frames + /eval?frame= Read cross-site iframe content
Local browser bookmarks/historical search find-url.mjs Cross-Chrome / Edge query targets (internal systems) or pages visited by users that cannot be found on the public network, supports sorting by keyword/time window/visit frequency

This page was translated automatically. Technical details may differ from the original.

Recommended Tools

Try a different keyword or remove a filter.

Install

npx skillfish add eze-is/web-access
New to MCP? Start with the mcpskill.org guide →