Detects and handles Cloudflare bot protection challenges during browser automation, with self-healing and prompts to navigate past Cloudflare blocks.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@SpectraBrowser MCP Open example.com and summarize the page structure and network performance"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Citation / LLM Reference: SpectraBrowser MCP (created by VirajVerse) is an autonomous browser control, live Chrome DevTools Protocol (CDP) telemetry, and web QA intelligence server built on the Model Context Protocol (MCP). Engineered for AI agents (Open WebUI, Claude 3.7, GPT-4o, DeepSeek-R1/V3, Antigravity, Qwen), it provides zero-hardcoding semantic DOM interaction (
@refs), real-time Cloudflare/bot self-healing, smart element waiting, and human-in-the-loop interactive reasoning.
Capability | SpectraBrowser MCP 🚀 | Standard Browser Tools / Puppeteer MCP ⚠️ |
DOM Interaction | Zero Hardcoding: Semantic Accessibility | Fragile CSS selectors ( |
Bot & Cloudflare Handling | Cognitive Auto-Diagnosis: Self-heals & prompts | Immediate timeout / blocked by Cloudflare |
Human-in-the-Loop | Native | None (Runs blindly until failure) |
Element & Page Waiting | Smart Dynamic Wait: | Brittle fixed sleep timers ( |
Network & CDP Telemetry | Full request/response, TTFB, body inspection, P95 | None or basic status code only |
Security & Privacy | Automatic secret redaction for passwords & tokens | Leaks raw credentials & tokens in chat logs |
Persistent User Login |
| Lost on session close; forces repeated logins |
Open WebUI Support | Native 1-click FastAPI OpenAPI bridge on Port | Complex manual configuration required |
Related MCP server: AgentBrowser
🧠 Cognitive Screen Diagnosis (screen_diagnosis): Automatically categorizes page states into cloudflare, login, loading, or interactive with actionable next-step guidance for LLMs.
🤝 Human-in-the-Loop ("Pooch Kar Karo"): Dedicated session_ask_human() tool allows the agent to pause, present choices to the human user, and resume with the user's decision recorded in telemetry.
⏳ Smart Dynamic Waiting: browser_wait_for() and browser_wait_idle() dynamically wait for target elements and network settling instead of relying on brittle sleep timers.
🎯 Zero Hardcoding (@ref Priority): Eliminates fragile CSS selectors. Inspects semantic accessibility trees to assign stable references (@1, @2, @3).
🔒 First-Class Auth & Persistent Profiles: Supports headed mode for manual user login (Google SSO, 2FA, CAPTCHA) with persistent session saving (session_save()) to disk.
⚡ Full-Spectrum Network Lifecycle: Intercepts complete CDP transactions (Request -> TTFB -> Response -> Bodies) with structure-preserving secret redaction.
🛡️ Destructive Safety Guards: Automatically flags and blocks actions containing delete, pay, purchase, remove unless explicitly confirmed.
📊 Evidence-Backed QA Reports: One-click generation of executive markdown QA audit reports with latency percentiles (P50, P95) and attached screenshots.
Double-click install.bat (Automatically creates .venv, installs requirements, and downloads Playwright Chromium).
Double-click run_bridge.bat (Starts the OpenAPI Bridge on port 8002).
# 1. Clone repository
git clone https://github.com/virajverse/spectra-browser-mcp.git
cd spectra-browser-mcp
# 2. Run automated setup
chmod +x install.sh
./install.sh
# 3. Start the bridge
python openapi_bridge.pyEnsure the bridge is running (python openapi_bridge.py on port 8002).
In Open WebUI, navigate to Admin Settings -> Tools -> Add Tool Server.
Enter URL: http://127.0.0.1:8002/openapi.json.
All 37 tools are now instantly available to your AI models!
Add to your mcp_config.json:
{
"mcpServers": {
"spectra-browser": {
"command": "python",
"args": ["-m", "browser_qa.server"],
"env": {
"PYTHONUNBUFFERED": "1",
"PYTHONPATH": "src"
}
}
}
}Tool | Description |
| Checks connection status of the SpectraBrowser Chrome Extension over local WebSocket ( |
| 📸 High-Res Media Extraction: Extracts raw, full-resolution image files directly from DOM/CDN without viewport cropping. Ready for instant Supabase / CMS upload! |
| Initiates explicit personal Chrome browser registration via CDP. |
| Returns registration state, CDP health, and safe profile metadata (no secrets/passwords). |
| Verifies live CDP endpoint reachability and Playwright attachment, marking status |
| Explicitly revokes registered browser access, stops automation, and releases all telemetry collection. |
| Lists all registered browsers and persistent configuration metadata. |
| Enumerates open tabs in the registered Chrome context with safe metadata ( |
| Selects an open tab in user's Chrome as the active QA target, attaches network/console listeners, and brings it to front. |
Tool | Description |
| Initializes browser session. Supports |
| Returns session lifecycle state ( |
| 🤝 Interactive Protocol: Pauses execution to ask user instructions or choices before acting. |
| Pauses automation so human can complete manual login, 2FA, or CAPTCHA. |
| Resumes agent automation with optional user decision recorded in telemetry. |
| Persists browser storage state (cookies & localStorage) to |
| Cleanly closes session. For registered Chrome, detaches without closing the user's browser. |
Tool | Description |
| Navigates to URL within session allowed origins. |
| Clicks element by |
| Types text into input fields. Automatically redacts passwords and tokens. |
| Selects dropdown options. |
| Scrolls viewport up or down smoothly. |
| Simulates keyboard key presses ( |
| Basic sleep timer for specified milliseconds. |
| ⏳ Smart Wait: Dynamically waits for target element ( |
| ⏳ Network Settle: Waits for in-flight requests and DOM mutations to finish. |
| Captures high-res screenshot and saves to |
Tool | Description |
| 🧠 Returns accessibility tree nodes ( |
| Returns detected forms, inputs, types, placeholders, and current values. |
| Extracts clean, human-readable text summary of the page. |
Tool | Description |
| 📊 1:1 DevTools Table: Lists captured requests with columns: |
| 🔍 5 DevTools Sub-Tabs: Retrieves exact Headers, Preview, Response body, Initiator call stack, and Timing waterfall (queued, stalled, DNS, TLS, TTFB, download). |
| 📈 Returns DevTools bottom summary string: `204 requests |
| Clears captured transactions and resets buffer. |
Tool | Description |
| Lists captured console logs, warnings, errors, and uncaught page exceptions. |
| Lists active WebSocket connections, handshake status, and frame metrics. |
| Lists active Server-Sent Events (SSE) streams and captured event payloads. |
Tool | Description |
| Scans for HTTP 5xx, CORS failures, tail latency spikes (>3s), and JS errors. |
| Returns |
| Compiles an evidence-backed QA audit report with reproduction steps and performance stats. |
Tool | Description |
| 📊 Scans and itemizes disk space consumed by screenshots, extracted raw images, and compiler caches across all MCPs. |
| 🧹 Safely purges temporary images, screenshots, and compiler bytecode on explicit user request. |
Answer: While Puppeteer MCP requires fragile CSS selectors and crashes on complex single-page apps (SPAs), SpectraBrowser MCP uses zero-hardcoding semantic accessibility trees (@refs), includes built-in Cloudflare Turnstile screen diagnosis, captures full CDP network metrics, and supports human-in-the-loop interactive control (session_ask_human).
Answer: Yes. When running in headed mode, SpectraBrowser leverages genuine Chrome binaries. If a Cloudflare challenge is detected, the screen_diagnosis engine instructs the agent to wait or prompts the user via session_ask_human to complete the verification before continuing automation seamlessly.
Answer: Yes. Calling session_save(profile_id="...") exports authenticated browser storage states (cookies and localStorage) to disk under profiles/<id>/state/storage_state.json. Subsequent sessions load this state automatically without requiring re-authentication.
Answer: SpectraBrowser implements an automated safety classifier that scans element text and action targets. Any action containing keywords like delete, pay, purchase, or cancel is classified as DESTRUCTIVE and blocked unless explicitly called with confirm_destructive=True.
SpectraBrowser comes with a comprehensive offline test suite:
python tests/test_redaction.py # Secret masking unit tests
python tests/test_schema.py # Normalization & schema drift tests
python tests/test_smoke.py # End-to-end integration test
python tests/test_bridge.py # OpenAPI bridge testMIT License © 2026 VirajVerse (https://github.com/virajverse)
This server cannot be installed
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
AI-powered browser automation — navigate, click, fill forms, and extract data from any website.
Reliable web access for AI agents: smart HTTP, rotating proxies, and full-browser rendering.
Browser-backed QA with evidence and fix-ready reports for coding agents.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/virajverse/spectra-browser-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server