Repolode
Submit
eze-is

eze-is/web-access

An Agent Skill that gives local AI agents like Claude Code full web access via tiered tool routing, a CDP browser proxy, and per-site experience notes.

View on GitHub
Stars
9,216
Forks
657
Open issues
57
Last commit
6 days ago

What is it?

web-access is an Agent Skill that adds web browsing and browser automation to AI agents that support SKILL.md, including Claude Code, Cursor, Gemini CLI and Codex CLI. It decides when to use built-in search, direct fetching, curl, Jina or a live browser, and it can run several sub-agents in parallel on multiple targets.

The browser part is a CDP proxy that connects directly to a Chromium-based browser the user already runs, such as Chrome, Edge, Brave or Arc, so existing logins carry over. It supports clicks, file uploads, screenshots, iframe access and extracting media from the page DOM. A separate script searches local bookmarks and history for pages that are not indexed publicly.

The skill stores notes on each site's URL patterns, platform quirks and known pitfalls, and reuses them across sessions. Browser preferences and site notes live in ~/.web-access/, outside the skill folder, so upgrades do not overwrite them.

Key features

  • 01Chooses among WebSearch, WebFetch, curl, Jina and CDP browser access based on the task.
  • 02Connects to a running local Chromium browser and reuses existing login sessions.
  • 03Supports JS clicks, real mouse clicks, coordinate clicks, file uploads and screenshots.
  • 04Reads content inside cross-site iframes and extracts media URLs or video frames.
  • 05Runs parallel sub-agents that share one proxy with per-tab isolation.
  • 06Stores per-domain site notes that persist across sessions.

Use cases

  • Researching several product websites in parallel and summarizing the differences.
  • Automating tasks on logged-in sites such as publishing posts from a creator dashboard.
  • Finding internal pages the user visited before but cannot remember the address of.
  • Letting a coding agent read dynamic pages that plain fetching cannot render.

Topics

#ai-agent#web-scraping#browser-automation#chrome-devtools-protocol#claude-code#agent-skill#web-search

FAQ

What is web-access?
An Agent Skill that gives local AI agents like Claude Code full web access via tiered tool routing, a CDP browser proxy, and per-site experience notes. web-access is an Agent Skill that adds web browsing and browser automation to AI agents that support SKILL.md, including Claude Code, Cursor, Gemini CLI and Codex CLI. It decides when to use built-in search, direct fetching, curl, Jina or a live browser, and it can run several sub-agents in parallel on multiple targets.
Is web-access free to use? What license does it have?
web-access has no declared license. Its code is public, but without a license you have no explicit permission to reuse it; ask the author before relying on it.
Is web-access still maintained?
Actively maintained. The last commit to web-access was on Oct 14, 2026.
What can I use web-access for?
Researching several product websites in parallel and summarizing the differences. Automating tasks on logged-in sites such as publishing posts from a creator dashboard. Finding internal pages the user visited before but cannot remember the address of. Letting a coding agent read dynamic pages that plain fetching cannot render.
Where can I get web-access?
The source code and installation instructions are on GitHub: https://github.com/eze-is/web-access

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects