Jev Browser
Headless browser automation via MCP server, CLI or library: Jev picks one action per step and scores goal-met and stuck probabilities.
Overview
Jev Browser drives a real headless browser from a task and a URL through an MCP server, CLI or library. At each step Jev picks one action from the page's clickable, typeable and selectable elements and scores how likely it is that the goal is met or the run is stuck, while code owns the loop, including budgets, recovery and stop gates. The result is the final page, a step trace with confidences, console errors and a screenshot.
The README lists real-site results: navigating Wikipedia from Coffee to Espresso in about 4 seconds for $0.0016, filling a contact form and stopping without submitting, pulling a price from a live pricing page, returning a guide as Markdown, and producing an accessibility-tree breakdown of a WordPress site. It is described as early software with rough edges on harder sites. It is published on npm as @jkudish/jev-browser, needs Node 22+, uses TypeSafe directly by default with alternative transports via the author's jev-agent-tools, optionally a typing model for text entry, and downloads Playwright Chromium on install. It has CI and an MIT licence.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which clickable, typeable or selectable element should be acted on next? | — |
| Yes / no | Is the goal met? | yes · no |
| Yes / no | Is the run stuck? | yes · no |
How it works
- Receive task and URL
- Extract actionable elements
- Jev picks an action and scores goal and stuck
- Code applies budgets, recovery and stop gates
- Return final page, trace and screenshot
Strengths
- Code-owned loop with explicit budgets
- Rich trace with confidences per step
- Very low cost per task on simple sites
Caveats
- Early software; harder sites may fail
- Text entry may need an extra typing model
Getting started
Register `npx -y @jkudish/jev-browser` as an MCP server and set TYPESAFE_API_KEY.At a glance
- Type
- MCP server
- Status
- beta
- Language
- TypeScript
- Runtime
- MCP server, CLI, library (Playwright)
- Has code
- yes
- Has tests
- yes
- License
- MIT
- Stars
- 335
Use cases and integrations
How curated lists describe it
Browser automation: drives a browser with Jev deciding each step, pitched as fast and very cheap next to LLM-driven browsing.
yibie/awesome-jevDrives a browser from a task and URL, returning the final page, screenshot and action trace.
logicrw/awesome-jev-projectsBrowser use using Typesafe's Jev model
wh000wh000/awesome-jev-liveBrowser-use experiment powered by Jev decisions.
cobanov/awesome-jevBrowser use using Typesafe's Jev model.
hellogumbo/awesome-jevHeadless browser automation through an MCP server, CLI, or library. Jev picks one action per step.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
The next action from the page's elements, plus probabilities that the goal is met or the run is stuck.
How do I use it from an agent?
Register the npm package as an MCP server.
Is it production-ready?
The README calls it early software.
Related projects
MCP server and libraries that drive Chrome over CDP and hand a decision model like Jev a bounded menu of page actions.
ProjectproductionTypeScriptClaude Desktop MCP extension that drives Brave or Chrome with Jev choosing on-page actions and a guard freezing irreversible clicks.
MCP serverbetaTypeScriptChrome MCP server with 50 tools that drives your everyday browser without losing logins, where Jev or a local decider picks each step.
MCP serverbetaStealth browser agent where Jev decides each step in about 0.3 s and CloakBrowser carries it out; MCP server, CLI and Python API.
MCP serverbetaPythonCodexQA browser skill that indexes in-page controls and has Jev pick the operation and target each step, with no screenshots needed.
Agent skillbetaAgent skill porting jev-ultrafast's loop into ego lite: one ~0.4 s Jev call per DOM step picks operation and target.
Agent skillbetaTypeScript