Resource Traversal MCP
Your agent’s key to the authenticated web
Kavia Shirkoohi
Try it out
Resource Traversal MCP is a custom Model Context Protocol server that gives your AI assistant authenticated access to the web – without a single API key. Using a hardware-locked Chromium profile stored locally on your machine, it inherits your existing browser sessions across Gemini, ChatGPT, Grok, Discord, Slack, Google Docs, NotebookLM, and more, then extracts their content through a modular library of bespoke DOM adapters. Powered by Playwright’s Native Headless Architecture, it masks all automation signatures to bypass bot detection, handles cookie consent walls and app redirects autonomously, and even walks back through dynamically loaded conversation histories to capture full temporal context – all privately, zero-cloud, with your data never leaving your machine. If you can see it in your browser, your AI can now read it.
FAQ
How can you use this?
A basic run-down for non-techies
The AI lives in your Terminal.
Instead of opening a browser tab to chat with an AI, some people run their AI directly from the Mac Terminal – a text-based window where you type commands. Tools like the Gemini CLI work exactly like that. You type a message, the AI replies, right there in the Terminal.
So where does Resource Traversal come in?
When the AI is running in the Terminal, it can’t just open a browser – it has no eyes. It only knows what you tell it or what tools it’s been given. Resource Traversal MCP is one of those tools. It runs quietly in the background on your Mac, and when the AI needs to look at a URL, it calls this tool, which spins up an invisible Chrome window, reads the page using your saved login session, and hands the content back to the AI – all within seconds, without you ever seeing it happen.
The full picture for a Mac user:
You open Terminal and chat with your AI (e.g., using the Gemini CLI)
You say something like: ”Read this Gemini chat and summarise it” and paste a link
Behind the scenes, the AI calls Resource Traversal MCP
An invisible Chrome window opens, logs in with your saved session, reads the page
The AI gets the full text back and responds to you – all in the same Terminal conversation
It’s like having a research assistant who can go and actually read anything on the web that you have access to, and report back to you while you stay in the conversation. You never leave the Terminal. You never touch the browser. It all just happens.
What makes this better than an AI-powered web browser?
Browser AI only sees what’s in front of it.
When you use AI in the browser – whether that’s a Chrome extension, a sidebar assistant, or even Antigravity running in the browser – it can only read the page you’re currently on. It’s essentially looking over your shoulder at one tab at a time. You’d have to manually navigate to each thing you want it to read.
Resource Traversal MCP lets the AI go and fetch things itself across services.
The difference is who’s doing the legwork. In the browser, you open the tab, you navigate to the right page, and then the AI reads it. With Resource Traversal, you just say ”here are five links – summarise all of them” and the AI fetches every single one (your Gemini chat, a Slack thread, a Google Doc) simultaneously, in the background, while you sit in one conversation.
It also means the AI can pull context you didn’t even think to mention. If it’s connected to Tana and sees a URL in your workspace, it can proactively go read it without you having to open anything.
The real edge is depth + reach combined.
Browser AI is reactive: it reads what you show it. Resource Traversal makes the AI proactive: it can go and gather its own context across your entire authenticated web, from one conversation, without you switching a single tab.
Think of it this way: browser AI is like having an assistant sitting next to you who can read the document on your desk. Resource Traversal is like having that same assistant who can also go to the filing cabinet, the group chat, and your colleague’s shared doc, and come back in under ten seconds.
How does it work?
For those who want to know more details
This application transforms static URL nodes from dead-end pointers into live cognitive portals.
It gives the AI “digital eyes and keys” to walk through the authenticated web as a first-class observer, enriching Tana’s context with data that is usually hidden behind login walls.
Unified Auth Persistence
Unlike conventional tools, this project utilises Native Chromium Stealth Architecture (
—headless=new ) to inherit the user’s actual browser permissions.One-Time Authentication: By leveraging a hardware-locked “Master Profile,” you only need to authenticate a service once (e.g., Gemini, ChatGPT, or a private Slack archive).
From that moment on, the AI inherits those sessions, bypassing the need for fragile API integrations, developer accounts, or permission collisions.
Technical Breakthroughs
Undetectable Traversal: Uses native masking to remove automation signatures (
navigator.webdriver), bypassing “nuclear-level” bot detection on high-security AI platforms.Temporal Navigation: Specialized logic enables the AI to “walk back” through deep chat histories (SPAs) to capture full conversational context.
Modular Extraction Framework: A plug-and-play architecture for site-specific extractors (Gemini, ChatGPT, Discord, Slack, etc.) with an intelligent Generic Fallback for everything else.
Security & Sovereignty
Local-First Architecture: User credentials and cookies never leave the machine.
Hardware-Locked Encryption: Sessions are secured via OS-level encryption (macOS Keychain / Windows DPAPI), physically anchoring the AI’s “keys” to the user’s hardware.
Consensual Observation: The AI only enters the portals the user explicitly unlocks, maintaining the highest ethical standards for agentic privacy.
Result: The Live Knowledge Graph
Transforms Tana from a static repository into a Live Knowledge Graph where every link is an active data-pipe, proactively expanding the AI’s memory and situational awareness.
Verified Cognitive Portals
AI Chats: Gemini, ChatGPT, Grok (Full history analysis).
Communities: Discord, Slack (Archives, Threads, and Real-time context).
Development: GitHub (Private/Public repositories and project structure).
Productivity: Google Docs, NotebookLM.
