Endoplexity
Agentic browser control driven by your own Claude / Cursor subscriptions
What is it?
What it is
A Chrome side‑panel extension that lets you run an existing AI CLI (Claude or Cursor) inside your real browser, providing click, type, navigate, file‑upload and other browser‑automation tools without needing a separate API key.
Why it exists
To give the AI loops you already pay for actual hands (a browser) and a face (a UI), eliminating per‑click billing and letting you use your real logged‑in session for tasks like form filling, price comparison, and document analysis.
Who should use it
Developers and power users who already have a Claude or Cursor subscription and want to let an AI agent control a real browser directly, without API costs. Ideal for automating form filling, page navigation, data extraction, and comparison tasks using the agent you already pay for. Comfortable with Node, Chrome extensions, and CLI usage.
Who should avoid it
Users without a Claude or Cursor subscription, those seeking a fully headless or turnkey SaaS solution, those needing macOS/Linux autostart (currently Windows‑first), and anyone uncomfortable with running a local bridge that interacts with a real Chrome instance and manages browser automation safety.
How it works
A quick walkthrough in plain English
How Endoplexity works
Step 1 of 3
You interact with it
Open Endoplexity, send a request, or connect it to your stack.
Features
Advantages
- No new API key required – uses already-paid AI subscriptions
- Uses your existing Claude or Cursor CLI as the agent loop
- Real browser session (MV3 Chrome debugger) rather than a headless copy
- Efficient input method – accessibility-tree snapshots reduce bandwidth vs raw HTML
- Single-turn action capability reduces token usage significantly
- Comprehensive security model with origin binding and token-gated access
- 156 unit tests plus an in-panel self-test for reliability
- Supports multiple AI models (Claude, Cursor) simultaneously
- Low cost – equivalent to $0.0959 per job application (already covered by subscription)
Disadvantages
- Windows-first implementation – macOS and Linux autostart not yet supported
- Session state stored in bridge memory – restarting loses resumability
- Real-world ATS comboboxes require special click-then-click flows (not standard <select> tools)
- Cannot drive chrome:// URLs or the Chrome Web Store due to browser restrictions
- After extensive navigation, dead iframe sessions may leave residual noise
- Trust mode disables safety gates entirely (fails closed otherwise)
- Some complex UI elements (non-select elements, unusual labels) may bypass the approval gate
Installation
native
git clone https://github.com/Endokelp/Endoplexity.git cd Endoplexity npm install npm run setup
FAQ
What is Endoplexity and how does it differ from other browser automation tools?
Endoplexity is a Chrome side‑panel that lets an AI model (Claude or Cursor) control your real browser session—clicking, typing, navigating, and filling forms—without a separate API key or metered billing. Unlike headless agents, it uses your existing browser, keeps your credentials on the machine, and exposes a small, safe set of browser tools over a local bridge. The panel owns the CDP connection, and the bridge enforces a strict safety policy, so the model never gains direct access to your accounts or the network.
What are the prerequisites to run Endoplexity?
You need Node 24+ (the bridge runs TypeScript directly), Chrome 114+ (the extension uses Chrome Debugging Protocol), and a logged‑in CLI for Claude or Cursor (or both). A Claude Pro plan is sufficient; choose Sonnet in the model dropdown. No API key is required.
How do I install, start the bridge, and load the extension?
Clone the repo and run `npm install`. Then start the bridge with `npm run setup` (creates an autostart launcher on Windows) or `npm start` to run it in the foreground. In Chrome, enable Developer mode, click Load unpacked, and select the `extension/` folder. The Endoplexity icon will appear in the toolbar.
How does Endoplexity prevent unwanted or irreversible actions?
The bridge implements an approval gate that blocks actions like submit, pay, delete, or confirm unless the user explicitly allows them. It matches the visible label of the element against a whitelist of English words. There are three autonomy modes—watch, normal, trust—chosen in the panel; the default is normal, which always enforces the gate. The bridge also runs the model in a sandboxed profile with strict permissions, and the WebSocket connection is origin‑pinned to the extension ID.
How can I attach files for the model to read or upload?
Drag a file onto the composer, paste it, or click the paperclip to add it as a chip. The bridge stores a copy and gives the model a key (e.g., `resume`) instead of a file path. The model can then use `read_file` to read the contents or `upload` to send it to a form field. Attachments survive bridge restarts, and you can also configure a `.endo-files.json` mapping to point at existing files.
Featured in Videos
YouTube tutorials and walkthroughs for Endoplexity
Alternatives
Similar projects ranked by category, topics, and text overlap.
