Endoplexity README

The project's README from github.com/Endokelp/Endoplexity. As one markdown file: README.md.

Endoplexity

Endoplexity

Agentic browser control, driven by the AI subscription you already pay for.

A Chrome side panel that reads and acts on the page you're actually looking at: click, type, fill forms, move between tabs, all through your existing claude or cursor-agent CLI instead of a metered API key.

Licence Node Tests API key

Endoplexity opening three pricing pages in separate tabs and comparing their free tiers in one table

Real time, no cuts beyond a trim and a speed-up · full-quality video

What you just watched

One instruction: "Open the pricing pages for Vercel, Netlify and Cloudflare Pages in three tabs, then compare their free tiers in one table."

It opened three tabs, read each pricing page, followed a plan link that actually lived on a different page, noticed when a link took it somewhere unexpected and recovered from it, then came back with a comparison table, the sources it used, and a takeaway naming the trade-off between Cloudflare's unlimited static bandwidth and Netlify's credit cap.

That banner reading "Endoplexity started debugging this browser" is Chrome's own, and it stays up the whole time. This is your real browser, your real session, on the pages you already had open. Not a headless copy that has to log back in to everything.

Why this exists

Most open-source "browser agent" projects need an API key and bill you per click. But if you already pay for Claude or Cursor, a tool-calling agent loop is sitting on your machine right now, fully paid for, doing nothing.

The brain didn't need building. What was missing was hands (a browser it can actually touch) and a face (a panel that shows its work). That's the whole project: a tool server and a UI. The agent loop is the CLI you already have installed.

How it works

The extension never talks to a model directly. The CLI never talks to Chrome. Neither one holds the other's credentials.

┌──────────────────────┐   WebSocket    ┌──────────────┐    MCP/HTTP   ┌──────────────────┐
│  Chrome side panel   │  origin-pinned │ local bridge │  token-gated  │  claude -p       │
│  (MV3)               │◄──────────────►│  127.0.0.1   │◄─────────────►│  cursor-agent -p │
│  owns chrome.debugger│                │  :8787       │               │  (the agent loop)│
└──────────────────────┘                └──────────────┘               └──────────────────┘

The panel owns the CDP connection itself rather than handing it to the service worker, which sidesteps MV3's idle teardown. The bridge exposes 14 browser tools over MCP, and it's where the safety policy actually lives. Never in a prompt, so nothing the model says can widen its own permissions.

Tools: snapshot navigate click type key upload read_file select scroll hover back forward tabs use_tab

Two decisions do most of the work here:

read_file reads an attached document as text (PDF, DOCX, XLSX, PPTX, CSV, JSON, Markdown, or any text file), so the agent can answer questions about an attachment instead of only pushing it into a form field. You attach one by dropping it on the composer — see Files.

Getting started

This takes about five minutes. No API key, no billing page, no dashboard to configure. Just Node, Chrome, and a CLI you probably already have logged in.

1. Check what you need first

2. Clone it and install

git clone https://github.com/Endokelp/Endoplexity.git
cd Endoplexity
npm install

3. Start the bridge

npm run setup

This writes a small launcher into your Windows Startup folder, so the bridge comes up automatically every time you log in. No console window stays open, no token to copy anywhere. If you'd rather run it by hand instead (or you're not on Windows), npm start runs it in the foreground so you can watch its log directly. If something ever seems stuck, .endo-bridge.log at the repo root is the first place to look.

4. Load the extension into Chrome

  1. Open chrome://extensions
  2. Turn on Developer mode (top right)
  3. Click Load unpacked
  4. Select the extension/ folder from this repo

You'll see the Endoplexity icon appear in your toolbar.

5. (Optional) Log in to Cursor too

Only needed if you want to run tasks through Cursor's models as well as Claude's:

npm run cursor-login

This is a one-time login into a bridge-owned Cursor profile, kept separate from your normal cursor-agent login so the two never collide.

6. Files

Drag a file onto the composer, paste one, or use the paperclip. It shows up as a chip, and from then on the agent can read it (read_file) or put it into a form field (upload). Any file type — anything it can't turn into text, it says so rather than guessing.

The bridge keeps its own copy and hands the model a key (resume), never a path. Attachments survive restarts, and the × on a chip deletes the copy.

If you'd rather point at a file where it already lives, .endo-files.json still works and behaves identically:

cp .endo-files.example.json .endo-files.json

{
  "resume": "C:\\Users\\you\\Documents\\resume.pdf"
}

Both files are gitignored, and so is the attachment directory — everything stays on your machine.

7. Run your first task

Click the Endoplexity icon to open the side panel, type what you want, and hit Run. A couple of things worth knowing right away:

Try something low-stakes first, like summarizing whatever page you're on before you point it at a form. For a deeper look at what each part of the panel shows and how the approval gate behaves, see docs/handrun.md.

The safety model

This drives a browser that's logged into your real accounts, so the boundaries here are worth stating plainly, including where they stop.

Found a hole? See SECURITY.md and please report it privately.

What it costs

A complete job application, filled agent-driven on a real Greenhouse form:
Cost $0.0959
Tokens 112,064
Turns 10
Wall clock 42s

That dollar figure is what the CLI reports as equivalent API spend. On a subscription, it's already covered by what you're paying monthly, which is the entire point of this project.

The same task measured flat against a 5-tool baseline, even while carrying about 18k tokens more of tool schema, so going from 5 tools to 13 cost nothing per run in practice. Page returns were later cut 3.8x by sending a full page once and then only the lines that changed afterward: 119,856 → 31,626 tokens on a ten-turn Hacker News task.

Status

Early. v0.0.1, and honest about it. The tool layer, approval gate, session continuity, and panel are all built and tested: 156 unit tests, plus an in-panel self-test that drives the real CDP layer against a cross-origin fixture (run await endo.selftest() in the panel's own console).

Known gaps, stated rather than hidden:

Development

npm test     # 156 tests, node:test, no framework, no build step
npm start    # run the bridge in the foreground to watch its log

There's no bundler and no framework here. The extension is plain HTML and JS, and the bridge is TypeScript run directly by Node 24.

docs/handrun.md is the manual verification checklist.

Licence

Apache-2.0. See NOTICE for trademark attribution.

Independent project. Not affiliated with, endorsed by, or sponsored by Perplexity AI, Anthropic, or Anysphere.

Endoplexity (Endokelp/Endoplexity)

cdp
Searching Database…

180 ·GitHub
Endoplexity
What is this?Endoplexity is a Chrome side-panel extension that lets an existing Claude or Cursor subscription drive your real browser — clicking, typing, filling forms, switching tabs — through the claude or cursor-agent CLI you already have logged in, instead of a separate metered API key.
Is this a browser extension, an app, or something else?It is two pieces working together: a Chrome side-panel extension that reads and acts on the page you're looking at, and a local Node bridge/tool server that talks to your claude or cursor-agent CLI over MCP. There is no cloud service or separate backend.
How do I get it running?Clone the repo and run the setup script to start the bridge (it auto-launches on Windows login), then open chrome://extensions, enable Developer mode, and 'Load unpacked' pointing at the extension/ folder from the repo.
Do I need an Anthropic API key to use it?No — it reuses the claude or cursor-agent CLI you already have logged into a real subscription (Claude Pro is enough). You just need Node 24+ and Chrome 114+ installed.
What are the known gaps right now?The project is v0.0.1 and Windows-first — macOS and Linux can run the bridge with npm start but don't have autostart yet. It also can't drive chrome:// pages or the Web Store, real ATS comboboxes need a click-then-click workaround instead of the select tool, and long navigation sessions can leave stray noise from dead iframe sessions.
Will this work the same on macOS or Linux, or is it really a Windows tool?The bridge itself is portable and runs anywhere with npm start, but the convenient one-command autostart installer currently only targets Windows's Startup folder.
Does any of my browsing or files get sent off my machine?No — the bridge only binds to 127.0.0.1 (loopback), never a public interface, and attachments/session files are gitignored and kept locally; nothing leaves your machine by design.
How does the approval gate decide what to block?It matches the clicked element's visible label against a short list of English words like 'submit', 'pay', 'delete', and 'confirm' before letting irreversible actions through. It's a label heuristic, not real page understanding, so icon-only buttons or unusual wording (e.g. 'Finish') can slip past it — watch mode exists for exactly that case.
What does running a task actually cost?A sample agent-driven job application came to $0.0959 in equivalent API spend across 112k tokens and 10 turns in 42 seconds — but since that's already covered by a Claude or Cursor subscription, there's nothing extra to pay.
Can it read and fill in attached files?Yes, drag a file onto the composer and the agent can read it as text (PDF, DOCX, XLSX, PPTX, CSV, JSON, Markdown, etc.) via a readfile tool, or push it into a form field via upload; attachments are kept locally and survive restarts.
Why build this instead of another browser-agent project?Most open-source browser agents require a separate metered API key and bill per click. Endoplexity instead reuses the tool-calling agent loop you're already paying for in Claude or Cursor, adding only the missing pieces: real CDP-based browser control and a panel to watch it work.
What are similar projects?bb-browser, XActions, mcp-chrome, BrowserOS, browser-harness, evenfire, holaOS, headroom
Which topics is it in?cdp, chrome, chrome-extension, browser-extension, browser-automation, model-context-protocol, cursor, mcp-server, claude, ai-agent, privacy, mcp
Creator
Endokelp
Size
4.8 MB
Stars
180
Updated
CommitAug 15
Age
2mo
Commits · 90 Days