Captures a visual representation of the current page viewport or entire page as a base64 encoded image ('full_page' defaults to false, capturing only the visible viewport; set true to capture the entire page). Side effects: none (read-only). Prerequisites: requires an active Chrome tab. Returns: base64 encoded image in specified format. Use this to visually verify UI state, layout, or rendering. Alternatives: 'inspect_dom' for raw HTML structure, 'get_performance_metrics' for rendering metrics.

Triggers a native mouse click event on a DOM element identified by CSS selector, executing click handlers and form submissions. Side effects: may trigger page navigation, form submission, or modify DOM state. Prerequisites: element must exist and be visible (getBoundingClientRect must return valid coordinates). Returns: success confirmation with click coordinates. Use this to interact with buttons, links, checkboxes. Alternatives: 'fill_input' for text input, 'evaluate_js' for complex interactions.

Focuses an input field via CSS selector and inserts text using native input simulation, triggering input/change events. Side effects: modifies DOM input value; triggers input/change event handlers. Prerequisites: element must exist, be visible, and be an input/textarea or contenteditable element. Returns: success confirmation. Use this to populate form fields, search boxes, text areas. Alternatives: 'evaluate_js' for direct value assignment without events, 'click_element' to focus manually.

Executes arbitrary JavaScript code in the page context, returning evaluated results as JSON-serializable values. Side effects: may modify DOM, state, or trigger network requests. Prerequisites: requires an active Chrome tab; script context must allow execution. Returns: JSON-serialized return value (or error if promise rejected). Use this for dynamic inspection, DOM manipulation, complex interactions. Alternatives: 'inspect_dom' for read-only DOM queries, 'click_element' for UI interactions.

Navigates the current Chrome tab to a specified URL, loading new page content. Side effects: destructive of current page state; discards unsaved work. Prerequisites: requires an active Chrome tab; URL must be valid and accessible (subject to local-only restrictions if enabled). Returns: navigation confirmation. Auth requirements: subject to same-origin policy; may require credentials for restricted URLs. Rate limits: none. Use this to change the current page. Alternatives: 'reload' to refresh current page.

Retrieves the complete HTML document or a contextual snippet around a search query. Side effects: none (read-only). Prerequisites: requires an active Chrome tab with loaded DOM. Returns: full HTML or snippet with context markers. Use this to inspect page structure, find elements by text, or verify rendering. Alternatives: 'evaluate_js' for complex DOM queries, 'capture_screenshot' for visual verification.

Enables the debugger and injects a breakpoint at the first statement of any script loaded after reloading the page. Side effects: reloads the current page (destructive of unsaved state). Prerequisites: requires an active Chrome tab. Returns: confirmation of debugger enablement and page reload. Use this to debug script execution from the page load. Alternatives: 'set_breakpoint' for targeting specific scripts/lines, 'pause_on_exceptions' for exception-based pausing.

Executes the current line of code without entering function calls, pausing at the next line. Side effects: advances debugger execution state. Prerequisites: requires an active, paused debugger session. Returns: confirmation of step execution. Use this to skip function internals during debugging. Alternatives: use 'resume' to continue full execution, 'step_over' enters functions, or 'evaluate_on_call_frame' to inspect state without stepping.

Continues full execution of the debugger from current breakpoint. Side effects: advances code execution until next breakpoint or completion. Prerequisites: requires an active, paused debugger session. Returns: confirmation of resume command. Use this to continue program flow after inspection. Alternatives: 'step_over' or 'step_out' for single-step execution.

Searches all cached script sources for a text pattern and returns matching locations (line and column numbers). Side effects: none (read-only query). Prerequisites: scripts must have been parsed and cached by the debugger. Returns: JSON array of matches with script ID, line/column numbers, and line preview. Use this to locate code before setting breakpoints. Alternatives: 'set_breakpoint' for direct breakpoint placement, 'evaluate_js' for runtime code discovery.

Sets a debugger breakpoint at a specific location, pausing execution when reached. Side effects: modifies debugger state (breakpoint added until removed). Prerequisites: requires an active Chrome tab; target script must be loaded. Returns: breakpoint identifier and location confirmation. Use this to debug specific code paths. Alternatives: 'pause_on_load' for early script execution, 'search_scripts' to find scripts by pattern.

Evaluates JavaScript expressions within the scope of a paused call frame, accessing local variables and call stack. Side effects: read-only by default; can modify state if expression includes mutations. Prerequisites: requires debugger to be paused at a breakpoint with active call frame. Returns: expression result with type and value. Use this to inspect variables and call stack during debugging. Alternatives: 'evaluate_js' for global scope evaluation, 'step_over' to advance without evaluation.

Reloads the current page, discarding all unsaved changes and re-fetching resources from the server. Side effects: destructive of unsaved state; clears dynamic DOM state. Prerequisites: requires an active Chrome tab. Returns: reload confirmation. Use this to refresh page content or reset to initial load state. Alternatives: 'navigate' to load a different URL, 'pause_on_load' to debug reload execution.

Scrolls the page by pixel offset, viewport pages, or to a specific element using CSS selector. Side effects: modifies DOM scroll position (observable but reversible). Prerequisites: requires an active Chrome tab with content. Returns: scroll completion confirmation. Use this to navigate within long pages or bring elements into view. Alternatives: 'click_element' to trigger scroll by clicking, 'evaluate_js' for custom scroll logic.

Removes a previously set debugger breakpoint by its ID, allowing execution to pass that location uninterrupted. Side effects: modifies debugger state (breakpoint deleted). Prerequisites: requires an active, paused debugger session with the breakpoint ID returned from 'set_breakpoint'. Returns: confirmation of breakpoint removal. Use this to clean up breakpoints or disable debugging at specific locations. Alternatives: 'set_breakpoint' to add new breakpoints, 'resume' to continue execution.

Stops and restarts the Chrome instance selected by 'instance_id' (the default instance when omitted) with remote debugging enabled, optionally configuring proxy and opt-in capability presets. Side effects: destructive - terminates that instance's Chrome process and all its open tabs; closes its debugging connection; other instances keep running. Prerequisites: requires CHROME_PATH environment variable or chrome in PATH. Returns: restart success confirmation listing the presets applied. Use this to reset browser state, apply proxy settings, enable experimental browser capabilities, recover from crashes. Alternatives: 'reload' to refresh page without restart, 'navigate' to load new content. Parameters: 'features' accepts a closed set - 'WEB_MCP' (enables the experimental WebMCP surface for sites that expose tools to the browser) or 'WEBGL_SOFTWARE' (forces SwiftShader software WebGL for GPU-less environments); arbitrary Chrome flags are not accepted.

Gracefully terminates the Chrome instance selected by 'instance_id' (the default instance when omitted) and closes all its debugging connections. Side effects: destructive - kills the Chrome process of that instance only; its open tabs are closed; unsaved data lost; other instances keep running. Prerequisites: the instance must be running. Returns: termination success confirmation. Use this to clean up resources, prevent zombie processes, or end a debugging session. Alternatives: 'restart_chrome' to restart instead of stop, 'close_instance' to also remove a secondary instance from the registry.

Opens a new independent Chrome instance. Side effects: launches a separate Chrome process with its own profile directory and remote-debugging port. Prerequisites: rejected when the server runs in --user-profile mode (Chrome's singleton profile lock). Returns: structured JSON with 'instance_id' (pass it as the 'instance_id' argument of other tools), 'host', 'port' and 'profile_dir'. Use this to isolate browsing sessions, cookies, proxies or WebMCP contexts from one another. Alternatives: 'open_tab' for additional tabs within an existing instance. Parameters: 'features' accepts a closed set - 'WEB_MCP' (enables the experimental WebMCP surface for sites that expose tools to the browser) or 'WEBGL_SOFTWARE' (forces SwiftShader software WebGL for GPU-less environments); 'headless' defaults to false (prefer false so the user can see the browser).

Lists all running or registered Chrome instances. Side effects: none (read-only registry snapshot). Returns: JSON array of instance descriptors with id, label, host, port, profile_dir, features and is_default. Use this to discover instance_ids before passing 'instance_id' to other tools. Alternatives: 'list_tabs' to enumerate tabs within an instance.

Closes and stops the specified Chrome instance by id. Side effects: destructive - for secondary instances the Chrome process is terminated and the instance is fully removed from the registry; for 'default' the process is stopped but the registry entry is kept and the instance is re-created lazily on the next tool call. Returns: structured JSON with the instance_id and whether it was removed from the registry. Use this to free resources used by sessions created with 'open_instance'. Alternatives: 'stop_chrome' to stop an instance, 'close_tab' to close a single tab.

Opens a new tab in the specified Chrome instance. Side effects: creates a browser tab and registers it under a generated Tab ID; the tab stays open until closed with 'close_tab'. Prerequisites: none - the instance is launched lazily if not running yet. Returns: structured JSON with 'tab_id' (use it as the 'tab_id' argument of other tools), 'target_id' (raw CDP target id), 'url' and 'label'. Use this to work with several pages in parallel while keeping their CDP state isolated per tab. Alternatives: 'navigate' to load content in the active tab, 'list_tabs' to enumerate already open tabs.

Lists all tabs currently open and registered in the specified Chrome instance. Side effects: none (read-only registry snapshot). Returns: structured JSON with 'active_tab_id', the current tab targeted by tools that omit 'tab_id', and a 'tabs' array of tab_id, label, target_id and url. When no tabs are registered, tools fall back to the instance's default single-tab connection and a 'note' explains it. Use this to discover Tab IDs before passing 'tab_id' to other tools. Alternatives: 'list_instances' to enumerate Chrome instances.

Closes a specific tab by its registered Tab ID. Side effects: destructive - removes the tab from the registry and closes it in Chrome; if it was the active tab, another registered tab becomes active. Prerequisites: the tab must exist (see 'list_tabs'). Returns: structured JSON with the closed 'tab_id' and the new 'active_tab_id' (null when no tabs remain). Use this to clean up tabs you no longer need. Alternatives: 'close_instance' to stop a whole Chrome instance.

Switches the active default tab to the specified Tab ID. Side effects: the switched tab becomes the default target of all subsequent tool calls in this instance that omit 'tab_id'; when 'activate' is true it is also brought to the foreground in the Chrome window ('activate' defaults to true). Prerequisites: the tab must exist (see 'list_tabs'). Returns: structured JSON with the new 'active_tab_id' and whether the tab was brought to the foreground. Use this before interacting with a different page without repeating 'tab_id' on every call. Alternatives: pass 'tab_id' directly on an individual tool call to address a tab without switching.

Retrieves intercepted HTTP/REST requests and WebSocket frames from network activity cache with filtering. Side effects: when 'clear' is true the cached requests and WebSocket frames are emptied after being returned. Prerequisites: requires active Chrome tab with network monitoring enabled. Returns: JSON array of requests/WebSocket frames with optional full details ('include_details' defaults to true; set false for summary only). Rate limits: none. Use this to audit API calls, debug network issues, inspect WebSocket traffic. Alternatives: browser DevTools Network tab, HAR file export.

Retrieves cached console messages including log, warning, error, info levels and uncaught exceptions. Side effects: when 'clear' is true the cached console messages are emptied after being returned. Prerequisites: requires an active Chrome tab. Returns: JSON array of console messages with timestamp, level, and text. Use this to debug script errors, monitor page health, inspect exception traces. Alternatives: browser DevTools Console, error logging services.

Captures runtime performance metrics including JS heap size, DOM node count, and layout timing. Side effects: none (read-only snapshot). Prerequisites: requires an active Chrome tab. Returns: JSON object mapping metric names to numeric values (e.g., JSHeapUsedSize, LayoutCount). Use this to monitor memory usage, detect memory leaks, or profile performance. Alternatives: 'profile_page_performance' for detailed tracing, browser DevTools Performance tab.

Records a performance trace of page execution, calculating Core Web Vitals (FCP, LCP, DCL, Load) and identifying Long Tasks (>50ms blocking). Side effects: may temporarily disable cache ('disable_cache' defaults to false); impacts page memory/CPU. Prerequisites: requires an active Chrome tab; trace recording uses background bandwidth. Returns: JSON with vitals, blocking time, and top 5 long tasks. Use this to optimize performance, identify bottlenecks, measure cold starts. Alternatives: browser DevTools Performance tab, real user monitoring (RUM).

Activates proxy authentication interception via Fetch domain, supplying credentials for authenticated proxies. Side effects: starts background event listener that auto-terminates after 30s inactivity; initiates pre-warming navigation. Auth requirements: username and password credentials required. Prerequisites: Chrome must be started with proxy-server flag. Rate limits: 30-second timeout per inactivity period. Returns: confirmation of proxy auth activation. Use this after 'restart_chrome' with proxy settings. Alternatives: manual proxy configuration in system settings.

EXPERIMENTAL: Sends raw Chrome DevTools Protocol (CDP) commands to the browser for advanced use cases not covered by specialized tools. Side effects: depends on command; may modify page state, DOM, or trigger navigation. Auth requirements: subject to local-only restrictions if enabled (Page.navigate checked). Prerequisites: requires knowledge of CDP protocol; active Chrome connection. Returns: raw CDP command response as JSON. Use this only when specialized tools inadequate. Alternatives: use domain-specific tools (navigate, click_element, evaluate_js).

EXPERIMENTAL: Retrieves unhandled CDP events from domains not covered by specialized listeners (network, console, etc.). Side effects: none (read-only cache access). Prerequisites: requires active Chrome connection with send_cdp_command or custom domain listeners active. Returns: JSON array of custom events with method, parameters, and timestamp. Use this to see Target, Debugger, or other domain events. Alternatives: domain-specific listeners (get_network_logs, get_console_logs).

Lists all WebMCP tools currently registered by the web page. Side effects: none (read-only state access). Prerequisites: WebMCP feature must be enabled in Chrome and the page must have registered tools. Returns: JSON array of available tools with schemas and frame IDs. Use this to discover capabilities exposed by websites implementing WebMCP.

Invokes a WebMCP tool registered by the current web page. Side effects: depends on the tool invoked; may modify page state, trigger network requests, or perform other actions defined by the page. Prerequisites: WebMCP feature must be enabled, target frame must exist, and the tool must be registered. Returns: The output of the tool invocation. Use this to interact with page-provided tools.

Returns the current status and result of a WebMCP tool invocation by its invocationId. Side effects: none (read-only state access). Prerequisites: the invocationId must have been returned by webmcp_invoke_tool (it is included in the timeout error when a page tool awaits user consent). Returns: JSON with toolName, frameId, input, status ('Pending', 'Completed', 'Error', or 'Canceled'), and output/errorText when available. Use this to poll a long-running invocation (e.g. one waiting for a human to approve a consent dialog on the page) without blocking. Alternatives: webmcp_list_invocations to see all invocations at once.

Lists all WebMCP tool invocations from this Chrome session with their current status. Side effects: none (read-only state access). Returns: JSON array with toolName, frameId, input, and status ('Pending', 'Completed', 'Error', or 'Canceled') for each invocation. Use this to find invocations that are waiting for user consent on the page (status 'Pending') or to recover an invocationId after losing it. Alternatives: webmcp_get_invocation for a single invocation by id.