CloakBrowser-Agent
Stealth browser agent where Jev decides each step in about 0.3 s and CloakBrowser carries it out; MCP server, CLI and Python API.
Overview
CloakBrowser Agent takes a goal in plain language: TypeSafe Jev decides every step, CloakBrowser (a stealth browser) carries it out like a human, and the result comes back as markdown. The README's example finds one-way flights from Zurich to London in 10 actions and 14 decisions, listing Jev's probability for each chosen target and the runners-up.
Each step is one request: Jev answers which operation and which element together using speculative fan-out (a target question per possible operation, using only the chosen one's answer). The README reports decisions averaging 0.28 to 0.43 s and about 1 s of decision time for a whole search task; a text model is called only when a field needs typing. When done, Jev also scores every section of the final page against the goal so only relevant sections are returned. It ships as an MCP server, CLI and Python API, and a free CloakBrowser licence key is recommended for the latest stealth build.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which operation and which element should be used next? | click · fill · ... with page elements |
| Score | How relevant is each section of the final page to the goal? | — |
How it works
- Receive a goal and starting URL
- Jev picks operation and element per step via fan-out
- CloakBrowser executes; a text model fills fields
- Jev ranks final page sections
- Return markdown with an action log and probabilities
Strengths
- Transparent probabilities and runners-up per step
- Fast decisions per the README
- Result filtered to relevant sections
Caveats
- Stealth browsing may violate some sites' terms
- Needs a CloakBrowser licence key for latest build
Getting started
Install the package, set your TypeSafe key, run cloakbrowser login for a free licence key, then call browse(goal=..., url=...).At a glance
- Type
- MCP server
- Status
- beta
- Language
- Python
- Runtime
- MCP server, CLI and Python API
- Has code
- yes
Use cases and integrations
How curated lists describe it
Jev-powered stealth browser agent. TypeSafe Jev decides each step in ~0.3 s, CloakBrowser carries it out like a human. MCP server, CLI and Python API.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
The operation and element for each step, then which page sections are relevant.
How fast?
About 0.28-0.43 s per decision in the authors' runs.
When is a text model used?
Only when a field needs typing.
Related projects
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 Claude 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 serverbetaCodexQA 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