Give your agent a browser that's cheaper, private, and yours.
Every other way to put an agent on the web either burns tokens on raw HTML, ships your session to someone's cloud, or locks you to one model. Pickle does none of that — it's a real desktop browser that runs on your machine and proves what it saves.
Local AI in the box · works with any MCP client · nothing leaves your device
Same agent. Same web. A fraction of the cost, and none of the exposure.
Pickle isn't a faster Chrome or a smarter chatbot. It's the missing layer between your agent and the web — and it's built around three things the alternatives make you give up.
Pages cost tokens. Pickle cuts the bill.
Agents normally swallow 30–60k tokens of raw HTML just to read one page. Pickle hands them a structured map — a small fraction of that — and shows the exact tokens saved on every step. On the bundled local model, that's $0 per run.
Your machine. Your logins. Your data.
Pickle runs on your computer with your real sessions — no cloud browser, no account, no page content shipped off to a server. A local model is in the box, so it works offline and nothing about what you browse leaves the device.
Any model. A window you control.
Bring any MCP-capable agent — you pick the brain, Pickle is the browser. It's a real window with native tabs you can watch and grab mid-task, and it stops for your approval before it buys anything.
The honest comparison.
There are four common ways to give an agent a browser today. Here's where each one leaves you — and where Pickle is different.
| Pickle | Chrome+ extension | Cloud AI browsere.g. hosted assistants | Headless automationPlaywright / scripts | |
|---|---|---|---|---|
| Token-efficient page reads | ✓ measured | ✗ raw HTML | ~ opaque | ✗ raw DOM |
| Runs on your machine, your logins | ✓ | ✓ | ✗ their cloud | ~ fresh profiles |
| Page content never leaves the device | ✓ | ~ per extension | ✗ | ~ |
| Use any model you want | ✓ any MCP | ~ | ✗ locked in | ✓ if you code it |
| You watch and take over live | ✓ native tabs | ✓ | ~ | ✗ invisible |
| Works with your agent, no code | ✓ | ~ | ✓ | ✗ code required |
| No per-use or subscription cloud fees | ✓ local | ✓ | ✗ subscription | ~ infra cost |
✓ built in · ~ partial or depends on setup · ✗ not without giving something up
Reading a page shouldn't cost 40,000 tokens.
A single modern web page is often 30–60k tokens of raw HTML. Pickle gives the agent a structured read — headings, links, and interactive elements with stable ids — that's a small fraction of that, and logs exactly what it saved. The savings aren't a slogan; they're a number in the toolbar.
Bring the agent you already use.
Pickle speaks the Model Context Protocol, so any MCP client — like Claude Desktop — can drive your real browser. Add one server entry and the agent gets a full set of browse, read, and shop tools.
- Install Pickle — the local runtime and MCP server come bundled.
- Add the Pickle server — one entry in your agent client's MCP config.
- Ask your agent to browse — it drives the tab you're watching.
{
"mcpServers": {
"pickle": {
"command": "pickle-mcp"
}
}
}
The app shows a copy-paste config on its Connect screen, matched to your client.
One cart across every store.
Pickle Search reads real product catalogs over the Universal Commerce Protocol when a store supports it, and clean web results when it doesn't. Your agent fills one cart across merchants; checkout splits it into a real cart at each store.
Download and hand your agent the keys.
Bundled with a local model so it works the moment it opens — no keys, no setup, offline-capable.
Windows 10/11 · ~1.3 GB (includes local AI) · early build