Vessel Browser

Click an element on the page by its index number, CSS selector, or visible text.

Close a browser tab by its ID.

Create a bookmark folder for organizing saved pages. Returns existing folder if the same name exists.

Capture the current browser session as a named checkpoint for later recovery.

Create a named folder for organizing bookmarks. If a folder with the same name already exists, return it instead of duplicating it.

Create a new tab group from the active tab or a specified tab. Optionally provide a name and color.

Open a new browser tab, optionally navigating to a URL.

Return the browser tab the human is actively looking at right now. Use this instead of list_tabs when you only need the focused tab.

Delete a previously saved named browser session.

Dismiss a modal, popup, newsletter gate, cookie banner, or blocking overlay using common close and decline actions.

Extract structured content from the current page. Modes: 'full' (default, everything), 'summary' (title+headings+stats), 'interactives_only' (clickable elements with indices), 'forms_only' (form fields only), 'text_only' (page text, no interactives), 'visible_only' (only currently visible, in-viewport, unobstructed elements plus active overlays), 'results_only' (likely primary search/result links only).

Return normalized structured data derived from page JSON-LD, microdata, RDFa, and high-signal meta tags. Useful for recipes, products, articles, events, FAQs, and other schema-rich pages.

Extract a table from the page as structured JSON rows with headers.

Extract the text content of a specific element by its index number or CSS selector.

Fill multiple form fields at once. Provide a map of field identifiers to values. Fields are matched by index, name, label, or placeholder. Much faster than calling type for each field individually.

Mark the current workflow step as done and move to the next one. Call this after completing each step.

Clear the active workflow tracker.

Begin tracking a multi-step web workflow. Vessel will show progress after every action so you always know where you are in the flow.

Check the current workflow progress.

Focus an input, button, or interactive element. Useful before pressing keys or to trigger focus-dependent UI.

Remove a folder. By default bookmarks in it are moved to Unsorted. Set delete_contents to true to delete them with the folder.

Rename an existing bookmark folder.

Go back in browser history.

Go forward in browser history.

Visually highlight an element or text on the page for the user. Use to draw attention to specific parts of the page. Highlights persist until cleared. Set persist=true to save the highlight so it re-appears when the user revisits this page.

Move the mouse pointer over an element to trigger hover states, tooltips, or dropdown menus.

Fill saved credentials into the active page's login form. Requires user consent. The password is filled directly into the page -- it is NEVER included in the response.

List saved human passwords for a domain, or all passwords. Returns metadata only (never passwords). Use human_vault_fill to fill credentials into a page. Requires user consent.

Delete a saved password. Requires user consent. This cannot be undone.

Inspect one element and its nearest local UI region by index, selector, or visible text.

List bookmark folders and saved pages. Optionally filter by folder name or ID.

List browser tab groups with names, colors, collapsed state, and member tab count.

List highlights related to the current browsing session. Includes saved persistent highlights plus the active tab's live text selection and any visible unsaved highlight marks. IMPORTANT: When the user says they highlighted or selected text, call this tool before falling back to screenshots or vision.

List previously saved named browser sessions with cookie and storage counts.

List all open browser tabs with their IDs, titles, and URLs.

Load a previously saved named session, restoring cookies, localStorage, and saved tabs.

Compound action: navigate to a login page, fill credentials, and submit. Handles the full login flow in one call.

Append markdown content to an existing note in the configured Obsidian vault.

Create a note for a bookmark or append bookmark details into an existing memory note.

List recent markdown notes in the configured Obsidian vault.

Write a markdown note into the configured Obsidian vault for research notes, breadcrumbs, or synthesis.

Capture the current page into the configured Obsidian vault as a markdown note with URL, excerpt, and content snapshot.

Search markdown notes in the configured Obsidian vault by title, path, body, and optional tags.

Show performance metrics: total tool calls, average duration, per-tool breakdown.

Navigate the active browser tab to a URL. Use postBody to submit data via POST request (e.g. form submissions).

Open a saved bookmark by its bookmark ID. Optionally open it in a new tab.

Move an existing bookmark or save the current page into a folder, creating the folder if needed.

Navigate to the next or previous page of results. Auto-detects pagination controls.

Press a keyboard key, optionally after focusing an element.

Publish or stream agent reasoning/status text into Vessel's in-browser transcript monitor. Intended for external harnesses that want to mirror live thinking into the browser UI.

Read the active tab's page content. Includes saved highlights plus any active text selection or visible unsaved highlights on the page. Supports scoped modes plus glance and debug.

Reload the current page.

Ungroup a tab. Defaults to the active tab.

Remove a persistent highlight by ID and clear it from any open tab. Use list_highlights to find IDs.

Restore a previously captured checkpoint by name or ID.

Save the current page, a specified URL, or a link target from the current page as a bookmark.

Persist the current cookies, localStorage, and tab layout under a reusable session name.

Capture a screenshot of the current page. Returns a base64-encoded PNG image.

Scroll the page up or down.

Scroll a specific element into view by index, selector, or visible text.

Compound action: find a search box on the current page, type a query, and submit. Returns the resulting page state.

Search bookmarks by title, URL, note, folder name, or folder summary.

Select an option in a dropdown by label or value.

Enable or disable ad blocking for the active tab or a matched tab. Reload after changes unless reload is false.

Change the color of a tab group.

Submit a form using a field index, submit button index, form selector, or button selector.

Analyze the current page and return the most relevant tools and suggested next actions. Call this when you're unsure what to do next — it reads the page context and tells you the optimal approach.

Switch to a different browser tab by ID or title/URL match.

Mark the current task as abandoned. Use this when the task cannot be completed or is no longer relevant.

Mark the task as blocked with a reason, or clear a blocker to resume. Use this when you are stuck and need human input to continue.

Add a note to the current task. Use this to record observations, intermediate results, or context for the human supervisor.

Mark the current task as completed. Optionally add a summary note. Use this when the task goal has been achieved.

Start tracking a task. Creates a task memory record with a goal that persists across actions and browser navigation. Use this at the beginning of a multi-step task so the human supervisor can see what you are working on.

Check the current task memory status including goal, progress, notes, and blocker.

Update the current task's next step or facts. Facts are merged with existing facts. Use this to record progress and keep the human supervisor informed.

Collapse or expand a tab group.

Type text into an input field or textarea. Clears existing content first.

Alias for type. Type text into an input field or textarea.

Undo the most recent agent action by restoring the browser to its state before that action ran. Works for click, type, submit, navigate, and similar mutating actions.

Fill a login form on the current page using stored credentials from the Agent Credential Vault. The credential values are filled directly into the page — they are NEVER returned in this response. The user will see a consent dialog before credentials are used.

Check whether stored credentials exist for a domain. Returns credential labels and usernames but NEVER password values. Use this before vault_login to verify credentials are available.

Generate a TOTP 2FA code from a stored secret and fill it into a code input field. The TOTP secret and generated code are NEVER returned — only filled directly into the page.

Clear the captured error buffer for the active tab.

Clear the captured console log buffer for the active tab.

Get console log entries captured from the active tab. Returns a rolling buffer of recent console.log, console.warn, console.error, etc. calls. Automatically starts capturing on first use.

Execute a JavaScript expression in the context of the active tab's page via the Runtime.evaluate CDP method. Supports async/await. This is a dangerous action — it can modify page state.

Get captured JavaScript errors and unhandled promise rejections from the active tab. Automatically starts capturing on first use.

Read browser storage for the active tab's origin. Supports localStorage, sessionStorage, cookies, and IndexedDB database listing.

Get computed CSS styles for an element matching a CSS selector. Optionally filter to specific properties.

Set or remove an HTML attribute on an element matching a CSS selector. This is a dangerous action that modifies the page.

Clear the captured network request buffer for the active tab.

Get captured network requests/responses from the active tab. Returns method, URL, status, timing, headers, and size. Automatically starts capturing on first use.

Get the response body for a specific network request by its request ID. Use vessel_devtools_network_log first to find the request ID.

Get a performance snapshot for the active tab including navigation timing, paint metrics, memory usage, and resource loading statistics.

Query the DOM of the active tab using a CSS selector. Returns matching elements with their attributes, node type, and optionally their HTML content. Limited to 50 results.

Set or remove a key in localStorage or sessionStorage for the active tab. This is a dangerous action that modifies page state.

Wait for text or a selector to appear on the current page.

Wait for the current page to finish loading after a click or form submission.

Search the open web using the configured default search engine.