js-reverse-mcp

Executes JavaScript code in a real Chrome/Chromium browser context through the Chrome DevTools Protocol (CDP), with built-in anti-detection mechanisms to evade bot-detection scripts. Implements stealth headers, user-agent spoofing, and WebDriver property masking to appear as legitimate browser traffic rather than automated tooling. Rebuilt from chrome-devtools-mcp architecture to optimize for AI agent workflows.

Automatically generates Model Context Protocol (MCP) tool definitions from browser capabilities, exposing Chrome DevTools operations as callable functions with strict JSON schemas. Handles parameter validation, return type serialization, and error mapping to MCP protocol standards. Enables AI agents to discover and invoke browser operations through standard MCP tool-calling interface without manual schema authoring.

Implements automatic retry logic for transient failures (network timeouts, element not found, navigation failures) with exponential backoff. Provides detailed error context (error type, stack trace, recovery action) for agent decision-making. Supports custom retry predicates for domain-specific failure handling. Distinguishes between recoverable and fatal errors.

Tracks browser performance metrics (page load time, JavaScript execution time, network latency) and resource usage (memory, CPU, network bandwidth). Provides performance data in structured format for agent analysis. Enables agents to make performance-aware decisions (skip slow pages, optimize workflows). Supports performance budgets and alerts.

Executes arbitrary JavaScript code within a real browser's JavaScript engine (V8 via Chrome), capturing return values, console output, and errors, then serializes results back to JSON for agent consumption. Handles async/await execution, Promise resolution, and complex object serialization. Provides execution timeout and memory limits to prevent runaway scripts from blocking the MCP server.

Provides high-level DOM query operations (select, find, filter) using CSS selectors, with built-in element interaction methods (click, type, scroll, hover). Abstracts low-level CDP commands into agent-friendly operations that return structured element metadata (text, attributes, position). Handles dynamic element waiting and stale element recovery.

Handles page navigation (goto, reload, back, forward) with configurable wait conditions (wait for load, network idle, specific elements). Tracks navigation history and page state, enabling agents to understand page transitions. Implements timeout handling for navigation failures and provides detailed navigation metadata (URL, title, load time).

Captures full-page or viewport screenshots as base64-encoded PNG/JPEG, with optional element highlighting and annotation. Provides visual feedback for agent workflows, enabling agents to understand page layout and validate visual state. Supports viewport size configuration and device emulation for responsive testing.

Intercepts HTTP/HTTPS requests at the network layer (via CDP Network domain), enabling agents to inspect, modify, or mock responses. Supports request filtering by URL pattern, method, and headers. Allows response injection (mock data, error responses) without modifying page code. Useful for testing error scenarios, bypassing API calls, or injecting test data.

Manages browser cookies and localStorage/sessionStorage with get/set/delete operations. Enables agents to persist authentication state, manage session tokens, and simulate user preferences. Supports cookie attributes (domain, path, expiry, secure, httpOnly) for realistic cookie handling. Useful for maintaining logged-in state across multiple page navigations.

Extracts page content (HTML, text, structured data) with optional parsing into structured formats (JSON, CSV). Supports CSS selector-based extraction, XPath queries, and JSON-LD/microdata parsing. Handles encoding detection and text normalization. Provides both raw HTML and parsed structured output for agent consumption.

Manages browser session lifecycle (create, reuse, destroy) with optional state persistence across restarts. Enables agents to maintain browser context (cookies, localStorage, history) between tool calls. Supports session pooling for concurrent agent tasks. Provides session metadata (creation time, last activity, resource usage) for monitoring.