Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

BrowserPaw

Chrome MCP server with 50 tools that drives your everyday browser without losing logins, where Jev or a local decider picks each step.

MCP serverbetaAGPL-3.0
View on GitHub ↗

Overview

BrowserPaw lets AI agents control your everyday Chrome browser through 50 MCP tools while keeping your logins, cookies and focus. The author built it to avoid the problems of sandbox-oriented browser automation: empty profiles that trigger 2FA, Windows profile locks, permission popups and heavy token use. It prunes the DOM tree by more than 85% (piercing Shadow DOM and iframes) and dispatches trusted CDP events on active sessions.

The chrome_act_toward_goal tool runs a local perceive, decide, act, verify micro-loop without an MCP round trip per step. Decisions come from TypeSafe Jev in the cloud, a custom endpoint, or per the listing a decider model downloaded to your machine, with a heuristic fallback. The loop is bounded (up to 60 steps in Jev mode, default 10, and 5 in heuristic mode), intercepts 14 destructive keywords such as pay, delete and submit, and escalates ambiguous or low-confidence steps to the calling agent. Docs are in English and Chinese.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich action should be taken next toward the goal on this page?—
Yes / noIs this step ambiguous or low-confidence enough to escalate?(inferred)yes · no

How it works

  1. Agent calls chrome_act_toward_goal with a goal
  2. Server perceives the pruned page
  3. Jev, a local decider or a heuristic picks the action
  4. Destructive keywords and low confidence escalate to the agent
  5. Act via CDP and verify; repeat within the step limit

Strengths

  • Works on real logged-in sessions without focus theft
  • Large token savings from DOM pruning
  • Local decider option

Caveats

  • Acting on real accounts carries risk; keyword gate is incomplete
  • AGPL-3.0

Getting started

See the Client Config and Releases pages; set TYPESAFE_API_KEY for remote Jev mode.

At a glance

Type
MCP server
Status
beta
Runtime
Chrome MCP server + extension
Has code
yes
License
AGPL-3.0

Use cases and integrations

How curated lists describe it

  • Drives your everyday Chrome from an agent across fifty MCP tools, with each step decided by Jev or by a decider model it downloads and runs on your own machine.

    MrJev/awesome-jev

Frequently asked questions

Does it use my real browser profile?

Yes, it controls your everyday Chrome with your logins.

Which models can decide steps?

TypeSafe Jev, a custom endpoint, a local decider or a heuristic fallback.

How does it limit risk?

Bounded steps, destructive-keyword interception and escalation of low-confidence steps.

Related projects

  • AskJev

    Claude Desktop MCP extension that drives Brave or Chrome with Jev choosing on-page actions and a guard freezing irreversible clicks.

    MCP serverbetaTypeScript
  • jev-browser

    Library, CLI and MCP server where an LLM states each step's outcome and Jev decides element, action, value and done/blocked/irreversible.

    MCP serverbetaTypeScript
  • jev-guard-mcp

    Experimental browser-only MCP server where Jev picks a code-owned action ID and a human approves before execution.

    MCP serverprototypeTypeScript
  • public-browser

    MCP server and libraries that drive Chrome over CDP and hand a decision model like Jev a bounded menu of page actions.

    ProjectproductionTypeScript
  • Jev Browser★ 335

    Headless browser automation via MCP server, CLI or library: Jev picks one action per step and scores goal-met and stuck probabilities.

    MCP serverbetaTypeScript
  • SystemOneHarness★ 193

    Agent harness that compiles an environment's finite actions into typed questions for Jev and gates each step by confidence.

    CLIbetaPython