Tools overview

All 43 tools by group, with the mode each needs.

Every page in this section is generated from the companion’s tool registrations by bun run docs:tools, so parameter names, types, enums and defaults match the running code. Each example is validated against the tool’s schema when the docs are generated.

Conventions shared by all tools:

• tabId is optional everywhere. Without it the companion uses the calling agent’s own tab or the only usable tab; see Tabs and sharing.
• ref values come from browser_snapshot or devtools_elements search; see Snapshots and refs.
• Search tools take query with optional regex: true, and paginate with offset and limit (default 50).
• Long results are clipped; ask for a narrower query, a specific id, or a higher limit.
• Files are written to ~/.browspark/artifacts unless a path is given.

Browser automation

Tool	Purpose
browser_status	Connection state, pairing instructions, usable tabs, open dialogs
browser_session	Launch, close, list and delete developer-browser contexts
browser_tabs	List, open, close, activate tabs across all windows
browser_navigate	Go to a URL, reload, back, forward
browser_snapshot	Accessible tree with refs, optionally as a diff
browser_read	Text, markdown, links, tables or HTML of the page or a ref
browser_fetch	Load a URL in a temporary tab and return its main content
browser_extract	Structured records from repeated elements
browser_click	Click, hover, drag
browser_fill	Replace the text of an input, textarea or contenteditable
browser_select	Choose options in a select
browser_upload	Set files on a file input
browser_key	Press keys or type text
browser_scroll	Scroll the page or a ref’s container
browser_screenshot	Viewport, full page or element image
browser_wait	Wait for text, URL or element state
browser_dialog	Accept or dismiss JavaScript dialogs
browser_batch	Several calls in one round trip
browser_policy	Allow or block domains per tab or by default
browser_download	List downloads or wait for one
browser_pdf	Print to PDF
browser_webmcp	List and invoke WebMCP tools a page exposes

Developer tools

Tool	Purpose	Mode
devtools_session	Start, stop, clear an inspection session	both
devtools_events	Read or wait for collected events	both
devtools_capabilities	Probe which protocol domains work for a tab	both
devtools_cdp	Raw DevTools Protocol command	developer
devtools_console	Search, inspect and wait on console output	both
devtools_evaluate	Run JavaScript in the page, a frame or a paused call frame	both
devtools_network	Search requests, bodies, frames, HAR, throttle, block, mock, replay	both
devtools_sources	Scripts, stylesheets, source maps, overrides	both
devtools_debugger	Breakpoints, stepping, scopes, watches	both
devtools_elements	DOM search, HTML and attribute edits, styles, box model, listeners, overlays	both
devtools_performance	Traces, Web Vitals, runtime metrics	both
devtools_profile	CPU sampling profiles	both
devtools_memory	Heap usage and growth (both); snapshots, diffs, retainers, allocation sampling (developer)	mixed
devtools_coverage	JavaScript and CSS coverage	both
devtools_storage	Cookies, web storage, IndexedDB, Cache Storage, quota, clearing	both
devtools_workers	Service workers and the web app manifest	mostly both
devtools_emulation	Devices, throttling, geolocation, media, locale, vision, animations	both
devtools_accessibility	Accessibility tree, audit, browser-reported issues	both
devtools_security	Connection state, certificate, mixed content	both
devtools_lighthouse	Lighthouse audit with reports	developer
devtools_recorder	Record, parameterise, replay, export to Playwright	both

The exact per-domain picture is in the capability matrix.