owl-browserlisted

# Owl Browser (agent rendering) Owl Browser is an AI-native browser. Instead of screenshots or raw HTML, it renders each page as **OwlMark**: a compact, handle-addressable text view of what is actually on screen. You **observe** the page, then **act on handles**. This is far cheaper than a screenshot and removes pixel-coordinate guessing. This plugin bundles the Owl Browser MCP server, so the tools below are available as tool calls (`browser_create_context`, `browser_navigate`, `browser_observe`, ...). They require `OWL_API_ENDPOINT` and `OWL_API_TOKEN` to point at a running Owl Browser instance (Docker or standalone). The same tools exist over REST at `POST $OWL_API_ENDPOINT/execute/<tool>`. ## The loop (do this, keep it short) ``` browser_create_context(render_mode="agent") -> browser_navigate(url) -> browser_observe # OwlMark text + handle table -> browser_click(handle) / browser_type(handle, text) -> browser_observe # see the result, repeat -> browser_close_context ``` - Call `browser_observe` after navigating and after every action. It is the only way you see the page. - Act using the handle tokens `observe` prints (e.g. `l5`, `b12`, `x27`). No CSS selectors, no pixel coordinates. - `browser_observe` blocks until the page is ready. Do not add a separate wait step. - Never screenshot to read text or find elements. Screenshot only to judge visual design or layout. ## Reading an OwlMark render `browser_observe` re