Skip to main content
Glama
obra

superpowers-chrome

by obra
OverviewSchemaSourceRelated ServersDiscussionsAdmin
JavaScript
Local

use_browser

Control a persistent Chrome browser to navigate, click, type, extract content, run scripts, and take screenshots, with each DOM action auto-capturing screenshots, markdown, HTML, and console logs.

Instructions

Control persistent Chrome browser with automatic page capture.

Every DOM action (navigate, click, type, select, eval) auto-captures to the session dir:

  • {prefix}.png — viewport screenshot

  • {prefix}.md — page content as structured markdown

  • {prefix}.html — full rendered DOM

  • {prefix}-console.txt — browser console messages

Prefer reading these files to using 'extract' or 'screenshot' whenever possible. Pages showing credential-shaped content (tokens, 2FA seeds) are never captured, and such values are redacted from all output.

Schema: 4 parameters — action, selector (CSS/XPath or null), payload (string or object), timeout (ms). selector targets a DOM element (null/omit for navigation, eval, tab management, etc.). payload is a string for simple actions (navigate=URL, type=text, eval=JS, keyboard_press=key). payload is an object for structured actions (set_viewport={width,height}, drag_drop={target}, etc.) — a JSON-encoded string of the same object works too. Tabs are tracked as sticky state; use switch_tab to change the active tab. Use action='help' for full per-action payload shapes.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
actionYesAction to perform. action='help' lists all actions with payload shapes.
payloadNoExtra data for the action. Both a plain object and an equivalent JSON-encoded string are accepted for every structured shape below (e.g. set_viewport accepts either {width:390,height:844} or '{"width":390,"height":844}'). Literal string for code/free-text actions, taken as-is and never JSON-parsed even if it happens to look like JSON (eval=JS source, type=literal text, await_text=literal text to wait for, select=literal option value). String or object for simple cases (navigate=URL, set_profile=name, new_tab=URL, attr=attribute name or {selector,attr}). Structured object (or its JSON-string equivalent) for the rest: set_viewport={width,height,mobile?} (no bare-string form), keyboard_press=key string or {key,modifiers:{shift?,ctrl?,alt?,meta?}}, extract=format string or {format:'text'|'html'|'markdown',selector?}, screenshot=path string or {path,fullpage?,selector?}, scroll=direction string or {deltaX?,deltaY?,selector?}, drag_drop=target selector string, {x,y} target coords, or {source,target}, mouse_move={x,y,steps?,fromX?,fromY?} (no bare-string form), file_upload=path string, JSON array-of-paths string, or {selector,files:[...]}, get_console_messages={since:epochMs} or a bare epoch-ms timestamp, switch_tab=tab index/url-substring/title-substring). See action='help' for per-action payload shapes.
timeoutNoTimeout in ms for await_element / await_text actions.
selectorNoCSS or XPath selector — what to act on. Null/omitted for actions that don't target an element (navigate, eval, list_tabs, etc.). XPath must start with / or //. dialog::accept and dialog::dismiss are special selectors for handling open dialogs.
tab_indexNoLegacy: behaves like switch_tab. Sets the active tab to this index before running the action. Prefer the switch_tab action.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changedv3.0.4
    • changedInput schema / properties / payload / description
      Previous value: -"Extra data for the action. String for simple cases (navigate=URL, type=text, eval=JS, keyboard_press=key, set_profile=name, new_tab=URL). Object for structured cases (set_viewport={width,height,mobile?}, keyboard_press={key,modifiers:{shift?,ctrl?,alt?,meta?}}, extract={format:'text'|'html'|'markdown'}, screenshot={path?,fullpage?}, scroll={deltaX?,deltaY?} or direction string, drag_drop={x,y} or selector string for target, mouse_move={x,y,steps?,fromX?,fromY?}, file_upload={files:[...]}, get_console_messages={since:epochMs}, await_text=text string or {text,timeout?}, switch_tab=tab index/url-substring/title-substring). See action='help' for per-action payload shapes."New value: +"Extra data for the action. Both a plain object and an equivalent JSON-encoded string are accepted for every structured shape below (e.g. set_viewport accepts either {width:390,height:844} or '{\"width\":390,\"height\":844}'). Literal string for code/free-text actions, taken as-is and never JSON-parsed even if it happens to look like JSON (eval=JS source, type=literal text, await_text=literal text to wait for, select=literal option value). String or object for simple cases (navigate=URL, set_profile=name, new_tab=URL, attr=attribute name or {selector,attr}). Structured object (or its JSON-string equivalent) for the rest: set_viewport={width,height,mobile?} (no bare-string form), keyboard_press=key string or {key,modifiers:{shift?,ctrl?,alt?,meta?}}, extract=format string or {format:'text'|'html'|'markdown',selector?}, screenshot=path string or {path,fullpage?,selector?}, scroll=direction string or {deltaX?,deltaY?,selector?}, drag_drop=target selector string, {x,y} target coords, or {source,target}, mouse_move={x,y,steps?,fromX?,fromY?} (no bare-string form), file_upload=path string, JSON array-of-paths string, or {selector,files:[...]}, get_console_messages={since:epochMs} or a bare epoch-ms timestamp, switch_tab=tab index/url-substring/title-substring). See action='help' for per-action payload shapes."
  2. First observedv3.0.1

TDQS

A4.3/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations only cover safety flags (readOnly=false, openWorld=true, idempotent=false). The description goes well beyond them: it enumerates the four artifacts written per action (.png/.md/.html/-console.txt), discloses the credential redaction/never-capture policy, and notes tabs are sticky state. That is exactly the side-effect context an agent needs before invoking.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loads the core purpose, then the capture artifacts in a scannable list before the parameter prose. Some of the payload-form text duplicates the schema and could be trimmed, but every section is readable and earns most of its space.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

With no output schema, 39 enum actions, and destructive-capable actions like kill_chrome/clear_cookies, the description does well by pointing to action='help' and explaining capture/redaction. It stops short of warning about irreversible actions or persistence across sessions, which would fully round it out.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the schema already documents action, payload, selector, and timeout forms. The description largely restates those same payload shapes rather than adding meaning; the only genuinely additive note is the sticky-tab pointer to switch_tab. Baseline 3 applies.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb and resource ('Control persistent Chrome browser') and immediately names the distinguishing capability: automatic page capture on every DOM action. An agent knows exactly what class of tool this is without opening the schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Gives an explicit preference rule — read the auto-captured files rather than calling 'extract' or 'screenshot' — and routes to action='help' for payload shapes. It lacks guidance on when not to use the tool or on the sticky-tab workflow's tradeoffs, but the core selection heuristic is present.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Deploy Server

Other Tools