SuperAPI AI Directory

Web Access

Agent Skill open-sourced by developer Yize Eze, thinking about how to access the internet

Open website

About this tool

What is Web Access

Web Access is an Agent Skill open-sourced by developer Yize Eze, designed specifically for Agent frameworks that support SKILL.md , such as Claude Code, Cursor, Gemini CLI, Codex CLI, etc. Through a three-layer architecture of networking strategy + CDP browser operation + site experience accumulation, it thoroughly solves the pain points of traditional Agents such as weak networking capabilities, poor browser automation, and inability to handle login states, letting Agents think about how to access the internet like real people, supporting concurrent execution across 100+ web pages simultaneously, with all operations completed in background tabs without disturbing the user's current work.

Web Access

Main Features of Web Access

  • Automatic selection of networking tools: WebSearch / WebFetch / curl / Jina / CDP, autonomously judged by scenario and freely combinable.
  • CDP Proxy browser operation: Directly connects to the user's everyday Chrome, naturally carrying login states, supporting dynamic pages, interactive operations, and video frame capture.
  • Three click methods:/click(JS click)、/clickAt(CDP real mouse events),/setFiles(file upload).
  • Local Chrome bookmark/history retrieval:find-url.mjs Locate URLs by keyword/time window/visit frequency, able to find internal systems that cannot be found via public search.
  • Parallel divide and conquer: Dispatch sub-Agents for parallel execution across multiple targets, sharing one Proxy, with tab-level isolation.
  • Site experience accumulation: Store operation experience by domain (URL patterns, platform characteristics, known pitfalls), reusable across sessions.
  • Media extraction: Directly obtain image/video URLs from the DOM, or capture and analyze frames at any point in a video.
  • DOM boundary penetration: eval recursive traversal can penetrate boundaries that selectors cannot cross, such as Shadow DOM and iframes.

How to use Web Access

  • Environment preparation: Ensure the system has Node.js 22+installed (below 22 requires additionally installing the ws module) and the Chrome browser.
  • Enable Chrome remote debugging: Enter chrome://inspect/#remote-debuggingin the Chrome address bar, check “Allow remote debugging for this browser instance”, and restart the browser if necessary.
  • One-click installation (recommended): Execute npx skills add eze-is/web-accessin the terminal, and the skills CLI will automatically detect the Agent environment and install to the correct location.
  • Agent automatic installation: Directly send instructions to the Agent 帮我安装这个 skill:https://github.com/eze-is/web-access。
  • Plugin installation (Claude Code): Execute claude plugin marketplace add https://github.com/eze-is/web-access then execute claude plugin install web-access@web-access --scope user。
  • Manual installation: Execute git clone https://github.com/eze-is/web-access ~/.claude/skills/web-access。
  • Environment self-check: The Agent automatically calls scripts/check-deps.mjs during runtime to complete pre-checks, no manual execution needed; if manual verification is required, run node "${CLAUDE_SKILL_DIR}/scripts/check-deps.mjs"。
  • Activate and use: After installation, enter 遵循 web-access skill or directly issue a networking task, such as "help me search for the latest developments on xxx", "go to Xiaohongshu and search for xxx's account".
  • CDP Proxy automatic management: The Agent automatically manages the Proxy lifecycle, directly connecting to Chrome via WebSocket and providing an HTTP API for page operations (creating tabs, executing JS, clicking, screenshots, scrolling, etc.).
  • Parallel divide-and-conquer usage: During multi-target research, the Agent automatically dispatches sub-agents to execute in parallel, sharing a single Proxy with tab-level isolation, and the main Agent receives only the aggregated results.
  • Disable Other Browser MCPs: It is recommended to disable other browser MCP services such as Chrome Devtools and Playwright MCP to prevent the model from "fighting itself".
  • Social Media Operation Risk Warning: Operating social platforms (such as Xiaohongshu, Weibo) through browser automation carries the risk of account throttling or banning,it is strongly recommended to use a secondary account for operations.

Web Access Official Website Address

  • Official Website Address:https://web-access.eze.is/
  • Github Repository:https://github.com/eze-is/web-access

Key Information and Usage Requirements for Web Access

  • Node.js 22+ is a hard requirement, the environment check script is located at scripts/check-deps.mjs
  • Chrome must have remote debugging enabled:chrome://inspect/#remote-debugging check allow remote debugging in
  • It is recommended to disable other browser MCP services(such as Playwright MCP, Chrome Devtools MCP) to prevent the model from "fighting itself"
  • Proxy automatically manages lifecycle: tabs created via /new are automatically closed after 15 minutes of inactivity, can be adjusted via the environment variable CDP_TAB_IDLE_TIMEOUT to adjust
  • Social Media Operation Risk Warning: Operating social platforms through browser automation carries the risk of account throttling or banning,it is strongly recommended to use a secondary account
  • Installation Path: Claude Code is located by default at ~/.claude/skills/web-access, the environment variable CLAUDE_SKILL_DIR is automatically set when the skill is loaded

Web Access Core Advantages

  • Built-in Login State: Reuses the user's everyday Chrome cookies, allowing access to platforms requiring login such as Xiaohongshu and Weibo without additional configuration.
  • Background Without Interference: All operations are performed in background tabs, without stealing screen focus, so users can use the browser normally.
  • Extremely Strong Concurrency: Supports operating 100+ web pages simultaneously, with 10 sub-agents conducting parallel research on different platforms.
  • Smarter the More You Use It: Site experience is automatically accumulated, and the next visit directly reuses verified effective URL patterns and interaction logic.
  • Social Media Automation: Supports automatically opening the platform, filling in copy, uploading images, and clicking to publish.
  • Cross-Platform Compatibility: Available on Windows / Linux / macOS (v2.4.1+ scripts have been migrated to Node.js).

Web Access Application Scenarios

  • Multi-Platform Public Opinion Research: 10 sub-agents simultaneously research platforms such as Xiaohongshu, Weibo, Bilibili, Boss Zhipin, GitHub, etc., each opening 10 tabs to scrape today's trending topics in parallel and summarize a report.
  • Accessing Content Behind Login: Reuses logged-in Xiaohongshu/Weibo accounts to directly search within the site, scroll to load, and extract post content and comment section content.
  • Information Authenticity Verification: Bypasses secondhand reports and directly accesses the original text on official websites/official platforms, avoiding the illusion of circular corroboration.
  • Automated Social Media Publishing: Automatically opens the platform web page, fills in copy, uploads images, and clicks to publish (note platform risk control).
  • Internal System Retrieval: Find internal platforms not searchable on the public internet or previously visited pages via local Chrome history/bookmarks.
  • Web automation testing: Supports manipulating video elements, form filling, CAPTCHA handling, and frame capture at any point in a video.

Web Access FAQ

Q: Which Agent frameworks does Web Access support?
A: Compatible with all Agents that support SKILL.md , including Claude Code, Cursor, Gemini CLI, Codex CLI, etc.
Q: Why is Chrome remote debugging needed? Can't we just use headless directly?
A: Connect to the user'severyday browservia Chrome DevTools Protocol, naturally carrying cookies and login state, able to handle dynamic pages that require login (such as Xiaohongshu, Weibo), and all operations happen in a background Tab without stealing screen focus.
Q: Will operating 100 web pages simultaneously cause lag?
A: All operations happen in background Tabs and do not interfere with foreground use. Proxy automatically tracks tabs created via /new , and automatically closes them after 15 minutes of inactivity to prevent orphan tabs.
Q: How is site experience stored and reused?
A: Automatically saves verified effective URL patterns, platform characteristics, known pitfalls, interaction logic, etc. by domain, reused across sessions, enabling a learning loop that gets smoother the more it is used.
Q: Can it be used on Windows?
A: Since v2.4.1, full cross-platform support has been available. Scripts have been migrated from bash to Node.js, and it can be used on Windows / Linux / macOS. v2.4.3 also fixed the path conversion error in Windows Git Bash.
Q: How do I search local Chrome bookmarks and history?
A: v2.5.0 adds scripts/find-url.mjs, supporting sorting by keyword, time window, and visit frequency. Typical scenarios include finding internal organizational systems (targets not searchable on the public internet) or looking up previously visited pages whose addresses you no longer remember.
Q: What API endpoints does CDP Proxy provide?
A: The main endpoints include:/new(create tab),/eval(execute JS),/click(JS click),/clickAt(real mouse click),/setFiles(file upload),/screenshot(screenshot),/scroll(scroll),/close(close tab),/health(status check).