OpticParse • PhishVision

The Autonomous Agent Web Browsing & Threat Shield Stack

Zero-CSS Visual Extraction • Zero-Day Phishing & Wallet Drainer Defense • MCP Server • Multi-Agent Toolkits

> ⭐ **Support Open Source:** If you find OpticParse or PhishVision useful, please **give us a Star on GitHub**! It helps us maintain free edge scrapers and datasets for everyone. --- ## ⚡ Feature & Architecture Comparison Looking for a **Firecrawl alternative**, **Crawl4AI alternative**, or a **Zero-Day Phishing & Visual Shield** for your autonomous AI agents? Here is how OpticParse + PhishVision compares to traditional scraping engines and legacy security blacklists: | Feature / Capability | OpticParse & PhishVision | Firecrawl | Crawl4AI | Jina Reader | Google Safe Browsing / VirusTotal | | :--- | :---: | :---: | :---: | :---: | :---: | | **Primary Extraction Engine** | **Multimodal AI Vision (Zero-CSS)** | HTML / Markdown | DOM Tree / XPath | Regex / Markdown | N/A (Security Only) | | **Visual Prompt Injection Shield (`ToxicCanvas`)** | 🛡️ **Built-in (Zero-Opacity & CSS Sanitization)** | ❌ None (Vulnerable) | ❌ None (Vulnerable) | ❌ None | N/A | | **0-Day Phishing & Wallet Drainer Shield** | 🛡️ **Built-in (Instant Day 0)** | ❌ None | ❌ None | ❌ None | ⚠️ Historical Blacklist (12-48h delay) | | **Machine Discovery Standard (A2A)** | ✅ **Native `agent.json` & `agent-service.json`** | ❌ None | ❌ None | ❌ None | ❌ None | | **In-IDE Developer Free Trial** | ✅ **200 Requests Out-of-the-Box** | ❌ Mandatory Credit Card | ❌ Self-hosted Setup | ⚠️ Limited Free Tier | ⚠️ Strict Rate Limits | | **In-Terminal Instant Checkout** | ✅ **ASCII QR in Console + .env Save**| ❌ Web Portal Only | ❌ None | ❌ Web Portal Only | ❌ Enterprise Contract | | **Agent Ecosystem Toolkits** | **LangChain, ElizaOS, AgentKit, MCP** | LangChain, LlamaIndex | Generic Python | Generic HTTP | None (Raw API) | | **Payment & Settlement Flexibility** | **USDC (Base/Polygon) / $0.01 M2M** | Stripe ($16+/mo SaaS) | None (Open Source) | Stripe ($$$) | Enterprise ($$$) | | **Token Optimization Noise Reduction** | **96% Token Reduction** | ~80% | ~70% | ~85% | N/A | --- ## 📌 Why This Stack Exists: The Dual-Engine Agent Architecture Giving an autonomous AI agent unrestricted internet access introduces three critical vulnerabilities: 1. **Context Window & DOM Fragility (OpticParse solves this):** Raw HTML consumes 90%+ of LLM token context, while rigid CSS/XPath selectors break the moment websites change frontend frameworks. 2. **Visual Prompt Injection & Screen Poisoning ([ToxicCanvas](https://opticparse.com/toxic-canvas.html) solves this):** Adversarial sites embed invisible CSS, zero-opacity layers, and prompt injection strings to hijack multimodal vision models (Claude Computer Use, GPT-4o, Operator). ToxicCanvas sanitizes DOM tokens before ingestion. 3. **The Autonomous Link Trap (PhishVision solves this):** When agents follow links autonomously, malicious actors exploit them with zero-day credential harvesting kits, fake Web3 dApps, and wallet drainers. **OpticParse + PhishVision provides the complete solution:** - **The Shield (PhishVision + ToxicCanvas):** Inspects target URLs in real-time for SSL age, brand spoofing, and visual injection attacks *before* the agent executes navigation. - **The Engine (OpticParse):** Visually renders authenticated pages at the edge, converting dynamic JavaScript into token-optimized clean Markdown (96% noise reduction) without brittle CSS selectors. - **Embodied AI & Edge Hardware Compatible:** Lightweight token streams structured for low-power edge processors (NVIDIA Jetson, embedded robotics telemetry) that cannot waste local battery and GPU cycles running headless browsers ([Learn more](https://opticparse.com/embodied-ai.html)). Operating **150 autonomous extraction & threat pipelines** across **13 industries**, the network auto-harvests **1,250+ verified intelligence records every 24 hours** at the global edge on Cloudflare Workers, R2, and D1. --- ## 🏛️ Autonomous Agent Dual-Shield Workflow ``` ┌─────────────────────────────────────────────────────────────┐ │ Autonomous AI Agent (LangChain / LlamaIndex) │ └──────────────────────────────┬──────────────────────────────┘ │ Target URL ▼ ┌──────────────────────────────┐ │ PhishVision Security Shield │ │ - Heuristic Brand Distance │ │ - Zero-Day Phishing Kits │ │ - Web3 Drainer Signatures │ └──────────────┬───────────────┘ │ ┌────────────────┴────────────────┐ │ Verdict │ [ MALICIOUS ] [ SAFE ] │ │ ▼ ▼ ┌───────────────────────────┐ ┌───────────────────────────┐ │ Halt & Shield Agent Loop │ │ OpticParse Edge Extractor │ │ (Protect API & Wallet) │ │ - Zero-CSS Visual Parse │ └───────────────────────────┘ │ - 96% Token Reduction │ └─────────────┬─────────────┘ │ Clean Markdown ▼ ┌───────────────────────────┐ │ LLM Context Window / RAG │ └───────────────────────────┘ ``` --- ## 🏛️ System Architecture ``` ┌─────────────────────────────────────────────────────────────────┐ │ OpticParse System Architecture │ ├─────────────────────────────────────────────────────────────────┤ │ │ │ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │ │ │ 150 Scraping │──▶│ Cloudflare │──▶│ Cloudflare │ │ │ │ Pipelines │ │ Workers Edge │ │ D1 + R2 Lake │ │ │ └──────────────┘ └──────────────┘ └──────┬───────┘ │ │ │ │ │ ┌───────────────────────────┼───────┐ │ │ │ Distribution │ │ │ │ ▼ ▼ ▼ ▼ │ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌─────┐ │ │ │ Kaggle │ │ Hugging │ │ RapidAPI │ │Ocean│ │ │ │ Hub │ │ Face Hub │ │ Gateway │ │ NFTs│ │ │ └──────────┘ └──────────┘ └──────────┘ └─────┘ │ │ │ │ ┌──────────────────────────────────────────────────────────┐ │ │ │ Anthropic Model Context Protocol (MCP) Server for Agents │ │ │ └──────────────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────────────┘ ``` --- ## 🎯 150-Pipeline Industry Coverage Matrix | Industry Vertical | Active Pipelines | Core Extractions & Capabilities | | :--- | :---: | :--- | | **🛡️ Cybersecurity & Threat Intel** | **70** | Zero-day phishing kits, typosquats, crypto-drainers, brand impersonation | | **🛒 E-Commerce & Retail Arbitrage** | **15** | Amazon undercut alerts, Shopify inventory snipers, price elasticity | | **💼 B2B Growth & Lead Signals** | **45** | Executive hiring telemetry, YC startup surge, wage trends, SEC filings | | **⚡ Finance, Crypto & DEX Arbitrage**| **20** | Liquidity pool spreads, token mint monitors, funding rate surveillance | --- ## 📊 Verified Adversarial Stress Benchmarks (Live September 2026 Audit) Both engines were independently benchmarked against live adversarial threats and enterprise-grade anti-bot defenses: ### 1. OpticParse: Zero-CSS Visual Extraction Stress Test Tested live on production backend against complex dynamic JavaScript, nested schemas, and anti-bot protected targets: | Target Page | Challenge Architecture | Result | Latency | Data Extracted | | :--- | :--- | :---: | :---: | :--- | | **Cloudflare Official Plans** | Enterprise Bot-Shield & obfuscated DOM | **100% PASS** | 42.1s | Extracted complete tiered pricing structure (`Free`, `Pro`, `Business`) | | **Stripe Global Pricing** | Heavy React hydration & dynamic cards | **100% PASS** | 38.2s | Extracted standard card processing fees (`2.9% + $0.30`) | | **Hacker News Frontpage** | Real-time community feed & score metadata | **100% PASS** | 42.7s | Extracted top 5 stories with real ranks and point totals | | **GitHub Trending Hub** | Nested repository schemas & live star counts | **100% PASS** | 32.3s | Extracted trending repositories (`skills`, `ponytail`) with exact stars | * **Zero-CSS Extraction Accuracy:** **4 / 4 (100.0% Success Rate)** * **DOM Fragility:** Zero broken CSS/XPath selectors; resilient to class renaming and layout refactors. > 💡 **Understanding the ~35s Latency: Why Multimodal Vision Beats Brittle 2-Second Scrapers** > * Traditional scrapers finish in 2 seconds because they only regex raw HTML text. The moment a site deploys dynamic React hydration, anti-bot protection, or changes CSS class names, **traditional scrapers break and return empty data or 403 Forbidden errors**. > * **OpticParse prioritizes zero-failure autonomous agent execution:** It spins up a full headless Chromium browser, bypasses anti-bot layers, captures a high-resolution visual buffer, and has a Multimodal Vision LLM reason over the page like a human. > * **Result:** 100% schema accuracy with **zero broken CSS selectors**, guaranteed for autonomous background agents where data integrity matters more than raw millisecond speed. --- ### 2. PhishVision: 0-Day Adversarial Threat & Crypto Drainer Audit Audited against **50 live zero-day malicious URLs** from the **OpenPhish global threat feed** (created within hours of test) + **20 difficult authentic authentication portals**: | Security Benchmark Metric | PhishVision Score | Traditional Baseline (Google Safe Browsing / VirusTotal on Day 0) | | :--- | :---: | :---: | | **0-Day Threat Catch Rate** | **68.0% (34 / 50)** | **~10% – 15%** (Fails because new domains lack historical reports) | | **Authentic Auth Portals (Clean)** | **95.0% (19 / 20)** | **>90%** (Industry standard for benign enterprise login portals) | | **Median Execution Latency** | **1.58 seconds** | **5 – 15 seconds** (Heavy commercial sandbox scanners) | #### Verified Live Threats Neutralized On First Contact: * 🛡️ **Uniswap Wallet Drainer:** `uniswap-interface.vercel.app` (BLOCKED) * 🛡️ **Trezor Hardware Seed Stealer:** `sso-trezor-com-start-x-auth.typedream.app` (BLOCKED) * 🛡️ **TrustWallet Crypto Harvester:** `trust-wallet-liart.vercel.app` (BLOCKED) * 🛡️ **Compromised WordPress Injections:** `bruceleephilosophy.com/texts/` (BLOCKED) * 🛡️ **Apple Brand Impersonator:** `apple-fruit.xyz` (BLOCKED) #### Verified Authentic Portals Cleared as Safe: * `accounts.google.com` • `github.com/login` • `dashboard.stripe.com/login` • `appleid.apple.com` • `reddit.com/login` • `auth.openai.com` (All CLEARED with zero false alarm interruptions). --- ## 🐍 Python & LangChain Quickstart (PyPI) Install the official Python SDK or LangChain multi-agent toolkit: ```bash # Core Python SDK (OpticParse AI Scraper + PhishVision) pip install opticparse-py # LangChain & CrewAI Tool Integration pip install langchain-opticparse # LlamaIndex Tool Integration pip install llama-index-tools-opticparse ``` ### LangChain Dual-Shield Agent Usage: ```python from langchain_opticparse import OpticParseTool, PhishVisionTool # Initialize tools phish_shield = PhishVisionTool(api_key="your_api_key") optic_scraper = OpticParseTool(api_key="your_api_key") target_url = "https://example.com/pricing" # 1. Pre-flight security audit: Block phishing kits & wallet drainers threat_audit = phish_shield.run({"url": target_url}) if threat_audit.get("verdict") == "MALICIOUS": print(f"🚨 Security Alert: Agent halted. Threat detected (Score: {threat_audit['threat_score']}/100)") else: # 2. Extract clean visual Markdown without brittle CSS selectors clean_data = optic_scraper.run({ "url": target_url, "query": "Extract tier prices, plan limits, and feature comparisons" }) print(clean_data) ``` > ⚡ **200 Free Trial Extractions & In-IDE Activation:** > Every IP receives **200 free trial requests** out of the box. When your free trial completes, the terminal automatically prints an ASCII QR code + 1-click checkout link. Upon paying on-chain ($10 Starter / $50 Growth / $200 Scale / $0.05 M2M via MetaMask or mobile wallet), your live API key is automatically minted, saved directly to your local `.env`, and your code resumes execution seamlessly. ### LlamaIndex Usage: ```python from llama_index.tools.opticparse import OpticParseToolSpec from llama_index.core.agent import FunctionCallingAgentWorker # Initialize tool spec and extract documents tool_spec = OpticParseToolSpec(api_key="your_api_key") docs = tool_spec.extract(url="https://example.com", query="Extract specs") print(docs[0].text) # Convert directly to agent tool list tools = tool_spec.to_tool_list() ``` ### Python Agent Usage: ```python from opticparse import OpticParse client = OpticParse(api_key="YOUR_API_KEY") # 1. Clean Markdown Extraction for LLM RAG pipelines res = client.extract_markdown("https://news.ycombinator.com") print(res["markdown"]) # 2. Autonomous Zero-Day Threat Inspection safety = client.detect_phishing("https://suspicious-dapp-claim.xyz") print(f"Verdict: {safety['verdict']} | Threat Score: {safety['threat_score']}/100") ``` --- ## 🤖 ElizaOS Autonomous Agent Plugin (npm) Install the official ElizaOS autonomous agent plugin: ```bash npm install opticparse-eliza-plugin ``` ### ElizaOS Usage: ```typescript import { AgentRuntime } from "@elizaos/core"; import { opticParsePlugin } from "opticparse-eliza-plugin"; export const webScoutAgent = { name: "WebScoutAgent", plugins: [opticParsePlugin], settings: { secrets: { OPTICPARSE_API_KEY: process.env.OPTICPARSE_API_KEY } } }; ``` --- ## 🤖 Model Context Protocol (MCP) Integration Connect **Claude Desktop, Cursor IDE, or AutoGen** directly to OpticParse in 1 click: ### 1. Install via Smithery ```bash npx -y @smithery/cli install @parastejpal987-cmyk/opticparse --client claude ``` ### 2. Manual Configuration (`claude_desktop_config.json`) ```json { "mcpServers": { "opticparse": { "command": "python", "args": ["-m", "mcp_server"], "env": { "OPTICPARSE_API_KEY": "YOUR_API_KEY" } } } } ``` ### 🛠️ Exposed AI Agent Tools: * `opticparse_scrape`: Vision-based structured data extraction from any web URL. * `phishvision_detect`: Real-time phishing and brand impersonation heuristic scanner. * `search_lessons`: Query indexed threat telemetry records. --- ## 📮 Postman Public API Collections Test the live endpoints instantly in Postman with **200 free trial requests** built-in: * **OpticParse Visual Web Scraper Collection:** [Run in Postman](https://god.net.objectstorage.softlayer.net/v1/AUTH_084e34cb-d467-422f-87d2-7c3a81283d08/postman-run-button/run-button.svg) — Includes visual extraction, token-optimized RAG markdown parsing, and live system health checks. * **PhishVision AI Threat Shield Collection:** Includes 0-day phishing scanning, brand impersonation detection, and permit2 drainer audits in <1.6s. * **Offline Import:** Both collections are available directly in this repository under [`postman/`](postman/): - `postman/opticparse_api_collection.json` - `postman/phishvision_api_collection.json` --- ## 🛡️ Embed Dynamic Scrapeability & Security Badges Showcase that your open-source repository, documentation, or SaaS application is scrapeable by autonomous AI agents or verified safe against zero-day phishing: ### 1. OpticParse Scrapeability Badge ```markdown [![Scrapeable with OpticParse](https://opticparse-mcp-portal.parastejpal987.workers.dev/api/badge/scrapeable?score=98)](https://opticparse.com) ``` Output: ![Scrapeable with OpticParse](https://opticparse-mcp-portal.parastejpal987.workers.dev/api/badge/scrapeable?score=98) ### 2. PhishVision 0-Day Safe Badge ```markdown [![PhishVision 0-Day Safe](https://opticparse-mcp-portal.parastejpal987.workers.dev/api/badge/phishvision?status=safe)](https://opticparse.com) ``` Output: ![PhishVision 0-Day Safe](https://opticparse-mcp-portal.parastejpal987.workers.dev/api/badge/phishvision?status=safe) --- ## 📦 Master Public Datasets (Hugging Face & Kaggle) All master datasets are public, verified, and streamable in **Apache Parquet & CSV format**: | Dataset Name | Records | Format | Direct Access | | :--- | :---: | :---: | :--- | | **Master 150-Template Catalog** | `150` | `CSV / Parquet` | [Hugging Face](https://huggingface.co/datasets/paras9909/opticparse-150-template-web-corpus) / [Kaggle](https://www.kaggle.com/datasets/parastejpal/opticparse-150-template-web-corpus) | | **PhishVision Threat Intelligence** | `70` | `CSV / Parquet` | [Hugging Face](https://huggingface.co/datasets/paras9909/opticparse-150-template-web-corpus) / [Kaggle](https://www.kaggle.com/datasets/parastejpal/opticparse-150-template-web-corpus) | | **E-Commerce & Retail Arbitrage** | `15` | `CSV / Parquet` | [Hugging Face](https://huggingface.co/datasets/paras9909/opticparse-150-template-web-corpus) / [Kaggle](https://www.kaggle.com/datasets/parastejpal/opticparse-150-template-web-corpus) | | **B2B Growth & Financial Signals** | `45` | `CSV / Parquet` | [Hugging Face](https://huggingface.co/datasets/paras9909/opticparse-150-template-web-corpus) / [Kaggle](https://www.kaggle.com/datasets/parastejpal/opticparse-150-template-web-corpus) | ### 💻 Load in Python (1-Line Quickstart): ```python import pandas as pd # Load master threat intelligence dataset df_threat = pd.read_csv("https://opticparse.com/threat_intel_dataset.csv") print(f"Loaded {len(df_threat)} live threat vectors") print(df_threat.head()) ``` --- ## ⚡ Commercial RapidAPI Gateway Pay-As-You-Go developer access with **sub-350ms global edge latency** at `$0.008/request`: 1. **`01-ai-web-scraper`**: Autonomous AI vision web scraper 2. **`02-dom-poisoner`**: Streaming adversarial anti-scraping tag injector 3. **`03-edge-proxy`**: Global unblockable fetcher proxy 4. **`04-rate-limit-bypasser`**: Residential edge IP rotator 5. **`05-realtime-sentiment`**: Live threat & telemetry database feed 👉 **Target Gateway URL:** `https://opticparse-rapidapi-gateway.parastejpal987.workers.dev` --- ## 🦊 Autonomous AI Agent Micropayments (HTTP 402 Machine Paywall) For autonomous bots, multi-agent frameworks (ElizaOS, AutoGPT, CrewAI), and automated scrapers with no human in the loop, OpticParse supports **instant on-chain settlement**: * **Price**: `$0.01 USDC` per request * **Supported Chains**: **Base**, **Polygon**, **Arbitrum** * **Treasury Address**: `0xd458E709e7d54fd3659EF66624A621Cde74EDD27` * **Machine Manifest**: [`https://opticparse.com/.well-known/agent.json`](https://opticparse.com/.well-known/agent.json) (Google A2A Standard) ### 🤖 1-Line Autonomous Agent Request: ```bash curl -X POST https://opticparse-edge.parastejpal987.workers.dev/api/edge/scrape \ -H "X-Payment-TxHash: " \ -H "Content-Type: application/json" \ -d '{"url": "https://news.ycombinator.com"}' ``` *No account creation, no credit card, and zero KYC required.* Every verified on-chain transfer is verified cryptographically via global RPC nodes and grants immediate edge execution. --- ## 🛡️ Interactive Threat DB Directory (21 Brands) Explore live brand safety evaluations and visual impersonation checks: * [Google Threat Dossier](https://opticparse.com/threat-db/google.com.html) * [PayPal Threat Dossier](https://opticparse.com/threat-db/paypal.com.html) * [Binance Threat Dossier](https://opticparse.com/threat-db/binance.com.html) * [Coinbase Threat Dossier](https://opticparse.com/threat-db/coinbase.com.html) * [Amazon Threat Dossier](https://opticparse.com/threat-db/amazon.com.html) * [Full Directory Index](https://opticparse.com/threat-db/) --- ## 🔍 Common Search Index & Developer Integrations OpticParse & PhishVision are natively integrated across the agent and web extraction ecosystem: * **Alternative To:** Firecrawl, Crawl4AI, Jina AI Reader, ScrapeGraphAI, Playwright Python, Puppeteer, Google Safe Browsing API, VirusTotal API, URLScan.io. * **Autonomous Agent Compatibility:** LangChain (`langchain-opticparse`), LlamaIndex (`llama-index-tools-opticparse`), ElizaOS (`opticparse-eliza-plugin`), Coinbase AgentKit (`opticparse-agentkit-action`), Model Context Protocol (`mcp_server.py`). * **Use Cases:** AI Web Scraping, Multimodal LLM Vision RAG, Zero-Day Phishing Detection, Web3 Crypto Drainer Defense, Real-Time SKU & Price Intelligence, Continuous Snappy Parquet Datasets. --- ## 🤝 Open Source & Licensing OpticParse developer tools, client SDKs, and MCP servers are proudly released under the **MIT License**. * **Author:** Paras Tejpal ([@parastejpal](https://github.com/parastejpal987-cmyk)) * **Official Website:** [https://opticparse.com](https://opticparse.com)