Asklear Browser MCP
OfficialThis server lets an MCP-compatible agent control a local Chrome browser through the Asklear extension while keeping page data and login state on your computer.
Open URLs in background, visible, or foreground tabs with
navigate.Snapshot a page’s accessibility tree with
observe.Click elements using snapshot and element references with
click.Type text into inputs with
fill.Scroll pages by a specified pixel amount with
scroll.Extract structured text or markdown from the current page with
extract.Isolate automation sessions using
session_idand correlate requests withrequest_id.
Provides tools to control a Chrome browser via the Asklear extension, enabling navigation, clicking, filling forms, scrolling, and extracting page content.
Click on "Deploy 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., "@Asklear Browser MCP Open example.com and extract the main page content"
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.
Asklear Browser MCP
Local browser collection gateway for Asklear — a stdio MCP server that drives your Chrome browser via the Asklear extension.
Page content and login state never leave your computer. All browser operations run locally; the Asklear service never sees your cookies, credentials, or page data.
What you get
Six browser tools, usable from any MCP-compatible agent (Claude Code, Codex, WorkBuddy, Trae, …):
Tool | Purpose |
| Open a URL in a background tab |
| Snapshot the current page's accessibility tree |
| Click an element |
| Type into an input |
| Scroll the page |
| Extract structured text/markdown from the page |
Related MCP server: deeporax-browser-mcp
Install
First, follow the illustrated Chrome extension installation guide using the same computer and Chrome profile as the Agent. The paired local connection requires extension 0.3.3 or newer. Agents helping with installation should proactively share the guide with the user. Matching packages are available here while the public download is being updated:
Choose the package for your environment, unzip it to a permanent folder, and load
the folder containing manifest.json via Chrome's chrome://extensions page with
Developer mode enabled.
The bundled extensions connect to http://127.0.0.1:8765; the Agent and
Connector must use that same address. These packages cannot be configured for a
different port at runtime. Editing config.json inside a downloaded package
does not change its compiled Connector address.
uv tool install git+https://github.com/Asklear/asklear-browser-mcp.gitOr, if you prefer pip:
pip install git+https://github.com/Asklear/asklear-browser-mcp.gitIf the package is already installed with uv, upgrade it and restart the existing background Connector so it runs the new version:
uv tool upgrade asklear-browser-mcp
asklear-browser-connector restartFor an installation made with pip, use pip install --upgrade git+https://github.com/Asklear/asklear-browser-mcp.git
instead of uv tool upgrade, then restart the Connector as above.
Download the current matching extension package above, replace the files in the
unpacked extension folder, and reload it in chrome://extensions before pairing.
Start the local service with asklear-browser-connector start before pairing.
Pair the extension once on the same computer. The user runs
asklear-browser-connector pair in a local terminal, opens the extension popup,
chooses Connect local service, and enters the displayed code. Keep the code
private; an Agent must not read, repeat, log, or transmit it. In the popup, wait
until the local service is connected before testing browser tools.
Configure
Add this to your MCP client configuration:
{
"mcpServers": {
"asklear-local": {
"command": "asklear-browser-agent",
"args": []
}
}
}Prerequisites
Google Chrome 116 or newer installed
macOS, Windows, and Linux are supported
The Asklear Chrome extension installed and enabled in the Chrome profile used by the Agent
Open the extension popup to check the browser and local service connection
Verify:
curl http://127.0.0.1:8765/health→{"extension_connected": true}; then usenavigateandextractto verify actual page extraction
How it works
Agent (Claude Code / Codex / …)
│ stdio JSON-RPC
▼
asklear-browser-agent (this package)
│ probe / call over loopback HTTP (127.0.0.1:8765)
▼
asklear-browser-connector (independent background daemon)
│ WebSocket
▼
Asklear Chrome extension → your Chrome tabThe first browser tool call starts the Connector in the background when it is
not already running. A second Agent reuses the same loopback daemon; its
session is isolated by session_id. When an Agent exits, the Connector stays
running for the next Agent.
The daemon keeps only local runtime state under
~/.asklear/browser-connector/:
connector.pid— current process IDconnector.log— daemon outputprocess-token-*— a private local process tokenextension-pairing-key— the private local pairing key
You normally do not need to manage it. For troubleshooting, the same package provides:
asklear-browser-connector start
asklear-browser-connector status
asklear-browser-connector stop
asklear-browser-connector restartstart and the automatic start path do not need an API key or OAuth. The
token is read from the private local state file and is never put in the
command line.
Two connections, one Asklear
This package provides browser collection only. For data queries (JD/Tmall/PDD/Douyin market data), use the hosted Asklear MCP:
Connection | Purpose | Transport |
| Data queries, market research | Remote HTTP (OAuth) |
| Browser collection, page extraction | Local stdio |
Both can coexist. Your agent picks the right tool automatically.
License
Apache-2.0. See LICENSE.
Available Tools
6 toolsclickC
Run the local Chrome browser click operation. Page results stay local to the Agent.
Inputs: session_id (required), element_ref (required), snapshot_id (required), request_id.
extractC
Run the local Chrome browser extract operation. Page results stay local to the Agent.
Inputs: session_id (required), request_id.
fillD
Run the local Chrome browser fill operation. Page results stay local to the Agent.
Inputs: value (required), session_id (required), element_ref (required), snapshot_id (required), +1 more.
observeC
Run the local Chrome browser observe operation. Page results stay local to the Agent.
Inputs: session_id (required), request_id.
scrollB
Run the local Chrome browser scroll operation. Page results stay local to the Agent.
Inputs: pixels (required), session_id (required), request_id.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 6 tools
Each tool name maps to a distinct browser action: navigate, click, fill, scroll, observe, and extract are mostly unambiguous. There is minor potential overlap between observe and extract since both involve reading page state, but their names and typical usage in browser automation are distinguishable.
All six tools follow the exact same single-word lowercase verb pattern. This is fully consistent and easy to predict.
Six tools is a well-scoped size for a browser automation server. Each tool covers a core interaction or observation action without unnecessary bloat.
The toolset covers the main browser workflow: navigate, observe/extract, click, fill, and scroll. Minor gaps such as explicit wait, back/forward navigation, reload, hover, or select operations are missing, but agents can work around most of these for common tasks.
Maintenance
Related MCP Connectors
Official Browserless MCP server: browser automation, scraping, screenshots and PDFs for AI agents.
Browser MCP for logged-in tasks. Uses your Chrome — credentials stay local. Zero-token replay.
Stealth web browser for agents: search, fetch, click, download and type in persistent MCP sessions.
Remote MCP server exposing your Node Command fleet (files, shell, processes, browser) to AI hosts.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceStandalone browser automation MCP server supporting both BrowserOS and standard Chrome. Enables 16 MCP tools for browser control including navigation, interaction, and content extraction.12 npmMIT
- AlicenseNot gradedqualityAmaintenanceMCP server to control a real Chrome browser from Claude, enabling navigation, DOM snapshots, clicks, typing, screenshots, and JS execution via a Chrome extension bridge.MIT
- AlicenseNot gradedqualityBmaintenanceEnables local Chrome control for AI clients over MCP stdio, HTTP, or CLI, exposing 20 tools to extract page text, number and click elements, type text, scroll, manage tabs, capture screenshots, and bind uploaded files to file inputs. Operations run entirely on the local machine with origin allowlists and refusal of untrusted URLs, without calling a cloud model or consuming model tokens.2,388 npmMIT
- AlicenseNot gradedqualityAmaintenanceA local MCP server that lets AI coding tools drive the user's real, already-logged-in Chrome — reading pages, clicking, typing, uploading files, taking screenshots, and inspecting tabs, cookies, console, and network traffic — via 25 browser automation tools that pass bot detection without a debugging port. Requires a companion Chrome extension and stays entirely on the user's machine.1 npmISC