autocli-skill: Web Platform Access for Claude Code via Chrome Sessions

What autocli-skill Solves for AI Agent Workflows

AI coding agents like Claude Code have no built-in access to authenticated platforms such as Twitter, Bilibili, Reddit, or Notion. Getting real-time data from these sources typically means registering for developer API access, managing credentials, and writing integration code. Many of these platforms restrict or charge for API access altogether.

autocli-skill takes a different approach. It wraps autocli, a Rust command-line tool that reaches 55+ platforms by reusing the user's existing Chrome login sessions. The agent calls a natural-language instruction, the skill translates it to the appropriate autocli command, executes the binary, and returns the results. No API keys are needed because authentication is handled by Chrome's existing cookies.

The skill is positioned as a companion for Claude Code, OpenClaw, and other AI agent runtimes. The README describes the underlying autocli binary as a complete Rust rewrite of an earlier tool called opencli, described as up to 12 times faster with significantly lower memory use.

Three Modes: Public, Browser, and Desktop

The skill groups platforms into three operational modes, each with different prerequisites.

Public-mode platforms require no browser and call public APIs directly. This group includes HackerNews, Dev.to, Lobsters, StackOverflow, Wikipedia, Arxiv, and BBC news. Fetching HackerNews top stories, for example, requires nothing beyond the installed binary:

autocli hackernews top --limit 20 --format json

Browser-mode platforms require Chrome to be open and the user to be logged in, plus the autocli Chrome extension installed from the project's GitHub releases. This group covers Twitter/X, Bilibili, Reddit, YouTube, Zhihu, Xiaohongshu, Weibo, Douban, Facebook, Instagram, TikTok, and others. Posting a tweet from the CLI looks like:

autocli twitter post --text "Hello from Claude!"

Desktop-mode platforms go further, requiring the target desktop application to be running. Cursor, Notion, ChatGPT desktop, Discord, and Codex fall into this group. The skill can send prompts to Cursor, read open files, or control conversation history in ChatGPT desktop.

The distinction matters because a failure in browser mode usually means Chrome is not open or the extension is missing, while a failure in desktop mode means the application itself has to be running.

Installing autocli-skill and the autocli Binary

Installation has two steps. First, the underlying autocli Rust binary must be installed separately; the README points to the nashsu/AutoCLI repository for instructions. Second, the skill itself can be installed by asking Claude naturally:

Help me install this skill: https://github.com/nashsu/AutoCLI-skill

Or manually via:

npx skills add https://github.com/nashsu/AutoCLI-skill

The README states that Claude Code must be restarted after installation for the skill to activate.

For browser-mode commands, the Chrome extension must also be installed from the latest release in the nashsu/AutoCLI repository. The README documents a troubleshooting table: `autocli: command not found` means the install script should be re-run and PATH verified; Chrome not being controlled means the extension is not loaded; login state not recognized means the user must log into the target site manually in Chrome first; browser command timeouts can be diagnosed with `autocli doctor`.

Platform Coverage and Command Reference

The README lists commands for each supported platform. Twitter/X offers 24 commands including trending, bookmarks, profile, post, reply, like, and follow. Bilibili offers 12, including hot, search, subtitle, and download. Reddit covers 15, including hot, frontpage, subreddit, upvote, save, and comment.

A sampling of cross-platform queries:

autocli bilibili hot --limit 10 --format json
autocli youtube search --query "LLM tutorial"
autocli reddit hot --subreddit MachineLearning
autocli yahoo-finance quote --symbol AAPL
autocli douban top250 --format json

The README describes how Claude handles these instructions automatically: given a natural-language request like "Search YouTube for LLM tutorials", Claude selects the right autocli command, executes it, and returns results in a formatted table with translated titles for non-English content. The translation step is handled by the Claude runtime, not the binary itself.

The 4.7MB binary size and zero runtime dependencies mean deployment on a new machine is straightforward: download the binary, add the extension, and the full platform surface is immediately accessible.

Limitations: Desktop Workflow Only, Fragile Sessions

The session-reuse architecture is both the feature and the constraint. It is convenient precisely because it avoids API registration, but it is fragile in ways that API keys are not. Chrome sessions expire. Platforms update their login flows and may invalidate existing sessions without warning. If a target site deploys bot-detection measures or changes its DOM structure, the relevant autocli commands will fail until the binary is updated.

The approach is fundamentally limited to desktop environments where Chrome is running and logged in. Server-side agents, CI pipelines, headless deployments, and Docker containers cannot use browser-mode or desktop-mode commands. Only the public-mode platforms (HackerNews, Wikipedia, Arxiv, etc.) are usable without a live browser.

The README does not document a rate limit policy or whether repeated automated requests to browser-mode platforms can trigger account restrictions. Users automating high-volume data pulls through an AI agent session should verify the terms of service of each target platform independently.

As of 2026-04-20, the repository had no declared licence. Reuse, modification, and commercial embedding are legally ambiguous without a licence.

Comparison with Browser Automation and Official APIs

The closest alternative is Playwright, a browser automation library. Playwright can also interact with authenticated Chrome sessions, but it requires writing explicit automation scripts in JavaScript, Python, or another supported language, and it does not translate natural language to commands. For a developer comfortable with Playwright, the autocli approach offers faster setup but less flexibility for custom workflows.

Official platform APIs, where they exist, are more reliable and less likely to break on platform updates, but they require application registration, key management, and often paid tiers for production-volume access. For casual or exploratory use cases, the session-reuse approach sidesteps all of that overhead. The trade-off is maintenance risk: any platform update can silently break the autocli command for that platform.

Editorial conclusion

autocli-skill is a practical fit for engineers who use Claude Code or a compatible AI agent framework daily and regularly need to pull or post data across social platforms, news aggregators, or desktop tools without maintaining a battery of API credentials. The hard requirement is a running Chrome instance, logged in to each target platform, with the autocli Chrome extension installed for browser-mode commands. Teams running headless or server-side agents get no value here; the architecture assumes a desktop workflow. The README does not list a licence, so legal reuse terms are unclear and should be confirmed before deploying this in a commercial product. The last push to the repository was on 2026-04-20.

Frequently asked questions

Official sources

1. Issues
2. nashsu/autocli-skill on GitHub
3. README