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

WindowsJev

Windows automation and durable research MCP server for Codex and Claude Code where Jev chooses among bounded UI Automation actions.

MCP serverbetaMIT
View on GitHub ↗

Overview

WindowsJev is a local MCP server that gives Codex or Claude Code one command to control Windows and run durable multi-source web research, returning a compact verified trace instead of streaming screenshots into the conversation. Simple browser searches and URL navigation use deterministic code and consume no TypeSafe tokens. General desktop tasks use Microsoft UI Automation to build a bounded action list, and Jev chooses among only those actions.

Tools include running a bounded Windows task, snapshotting actionable controls without AI, and a research job system that starts independent workers, checkpoints progress, supports cancel and retry, and lets the host read or search artifacts locally. A rank tool spends one batched Jev call to rerank a deterministic shortlist of evidence. It is MIT licensed with CI and releases.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich of the bounded UI Automation actions should be taken next?—
ScoreHow relevant is each shortlisted piece of research evidence to the objective?—

How it works

  1. The host LLM calls run_windows_task with a goal.
  2. Deterministic code handles simple navigation; UI Automation lists actions otherwise.
  3. Jev chooses among bounded actions until done.
  4. A compact verified trace is returned.

Strengths

  • One tool call instead of a model turn per click.
  • Zero Jev tokens for simple navigation.
  • Durable, checkpointed research jobs.

Caveats

  • Windows only.
  • Jev choices limited to what UI Automation exposes.

At a glance

Type
MCP server
Status
beta
Runtime
MCP server (Windows)
Has code
yes
Has tests
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Token-efficient Windows automation and durable research MCP server for Codex and Claude Code, powered by TypeSafe Jev.

    logicrw/awesome-jev-projects

  • Token-efficient Windows automation and durable research MCP server for Codex and Claude Code, powered by TypeSafe Jev.

    hellogumbo/awesome-jev

Frequently asked questions

Does it send screenshots to the LLM?

No, it returns a compact trace.

When is Jev used?

To choose among bounded UI Automation actions and to rerank research evidence.

Which agents does it support?

Codex and Claude Code.

Related projects

  • arc-cua

    macOS computer-use package: a background app driver for any agent, plus a loop where Jev picks each UI action for a bounded subtask.

    ProjectbetaPython
  • Jev-Desktop

    Windows MCP server and CLI where Jev observes windows and chooses controls so agents can hand off whole desktop tasks.

    MCP serverbetaPython
  • macos-computer-use-kit

    Accessibility-first macOS computer-use MCP server and CLI, with optional Jev guards on targets and inputs before irreversible actions.

    PluginbetaPython
  • reflex-cu

    Computer-use MCP server for macOS and Windows that reads the UI tree and uses Jev to find elements, check screens and wait for states.

    MCP serverbetaPython
  • jev-browser-use

    Local browser automation workbench with an Edge/Chrome sidebar and Codex MCP sharing one Jev-driven task engine, with user confirmation.

    MCP serverbetaTypeScript
  • Jev-cu

    Codex computer-use skill where Jev picks the next on-screen element, action, completion and risk from text candidates.

    Agent skillprototypeJavaScript