sitewise

Control the web from your terminal. Sites as commands. Browser automation without an extension. A native MCP server for AI agents.

sitewise hackernews top --limit 5
sitewise reddit hot programming
sitewise read https://any-article.com --json
sitewise browser goto github.com

Why sitewise

• Zero setup. npm i -g sitewise and the core commands work immediately — no extension, no daemon, no account. Works on servers and in CI.
• Agent-native. sitewise mcp runs a full MCP server over stdio. Add it to Claude Code or Cursor and your agent gets site commands, article extraction, and browser control as first-class tools.
• Real browser control when you need it. sitewise browser drives a Chrome-family browser directly over the DevTools protocol — navigate, click, fill, extract, screenshot — using a dedicated profile where your logins persist.
• Human and machine output. Beautiful tables in a terminal; clean JSON with --json or whenever output is piped.

Install

Requires Node.js >= 20.

npm install -g sitewise
sitewise doctor

Site commands

51 sites and counting. Run sitewise list to see them all with their mode.

API adapters — instant, no browser, work anywhere (including CI):

| Category | Sites | |----------|-------| | Dev & packages | github gitlab npm pypi crates stackoverflow devto | | News & knowledge | hackernews reddit arxiv wikipedia wikidata rss | | Books & words | openlibrary gutenberg dictionary urban words quotes | | Media | itunes tvmaze anime musicbrainz rickandmorty pokemon | | Finance & crypto | crypto coinpaprika coinlore defillama trending fx | | Food, world & fun | cocktail meal brewery weather ip nasa worldtime art trivia joke lyrics |

Browser adapters — run through your logged-in browser for sites with no usable API:

| Site | Commands | |------|----------| | youtube | search | | githubtrending | today | | producthunt | today | | lobsters | hot | | steam | search | | imdb | search | | amazon | search | | twitter | search, profile (needs login) | | linkedin | search (needs login) |

sitewise npm search "web framework" -l 5
sitewise github search "language:rust http server" -l 5
sitewise arxiv recent cs.AI
sitewise weather now "San Francisco"
sitewise youtube search "rust async" -l 5
sitewise twitter profile jack           # uses your logged-in X session
sitewise list                           # everything, with mode (api/browser)

Every command supports --json for structured output.

Write your own adapter (no engine code)

Browser adapters are just data: a URL, an optional wait selector, and a set of CSS field selectors. Drop a YAML file in ~/.sitewise/adapters/ and it shows up in sitewise list:

name: news
description: "[browser] Example news site"
commands:
  - name: top
    description: Top headlines
    url: https://example.com/news
    waitFor: article.story
    columns: [headline, section]
    list:
      selector: article.story
      limit: 20
      fields:
        headline: { selector: h2 }
        section: { selector: .section }
        url: { selector: a, attr: href }

Because it's plain data, selectors that drift can be fixed in seconds — and AI agents can author new adapters through the MCP server.

Read anything as Markdown

sitewise read https://example.com/some-article

Extracts the main article (title, byline, body) and prints clean Markdown. Great for piping into an LLM.

Browser control

sitewise browser start                 # launches Chrome with a dedicated sitewise profile
sitewise browser goto news.ycombinator.com
sitewise browser text                  # visible text of the page
sitewise browser read                  # main article as Markdown
sitewise browser click 'a.morelink'
sitewise browser fill 'input[name=q]' "query"
sitewise browser screenshot -o page.png
sitewise browser tabs                  # list tabs; target one with -t <index|substring>

Log in to sites once in the sitewise browser window — sessions persist in ~/.sitewise/browser-profile.

For AI agents (MCP)

Add sitewise as an MCP server:

claude mcp add sitewise -- sitewise mcp

The agent gets tools for listing/running site commands (run_command), reading URLs as Markdown (read_url), and full browser control (browser_goto, browser_click, browser_fill, browser_text, browser_read, browser_eval).

License

MIT © Mehal Srivastava