Jev-Desktop
Windows MCP server and CLI where Jev observes windows and chooses controls so agents can hand off whole desktop tasks.
Overview
Jev-Desktop lets AI agents use Windows applications through MCP or the command line. An agent hands off a goal in one desktop_run call, and Jev observes windows, chooses controls, types supplied text and navigates locally; the agent gets back a result or a request for help rather than handling every click.
It needs Windows 10 or 11 with an interactive session and Python 3.12+. A TypeSafe key is required only when Jev selects controls; inspection and direct actions work without one. It installs as a Python package with a doctor command, and ships plugin manifests for Codex and Claude Code plus a desktop-use skill with operating instructions. Example tasks include using Calculator's buttons to compute a sum and opening a URL through Chrome's address bar.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which window control should be used next for this goal?(inferred) | — |
| Yes / no | Is the goal complete or is help needed?(inferred) | — |
How it works
- Agent discovers the target app and calls desktop_run with a goal
- Jev Desktop observes the window's controls
- Jev selects controls and the broker types supplied text and navigates
- Return the result or a help request to the agent
Strengths
- Agent hands off whole tasks instead of each click
- Inspection and direct actions work without a key
- Plugins for Codex and Claude Code
Caveats
- Windows only with an interactive desktop
- Does not auto-load dotenv; key must be in the client environment
Getting started
In a venv: `python -m pip install -e .` then `jev-desktop doctor`; add the MCP server with `python -m jev_desktop.transports.mcp_stdio`.At a glance
- Type
- MCP server
- Status
- beta
- Language
- Python
- Runtime
- MCP server and CLI on Windows
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Jev Desktop lets AI agents use Windows applications through MCP or the command line. Hand off a task and let Jev observe windows, choose controls, type supplied text, and navigate locally. Your agent gets the result or a request for help instead of handling every click.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
Which window controls to use to reach the goal.
Do I always need a TypeSafe key?
Only when Jev selects controls; inspection and direct actions work without one.
Which agents are supported?
Any MCP client, with plugin manifests for Codex and Claude Code.
Related projects
macOS computer-use package: a background app driver for any agent, plus a loop where Jev picks each UI action for a bounded subtask.
ProjectbetaPythonAccessibility-first macOS computer-use MCP server and CLI, with optional Jev guards on targets and inputs before irreversible actions.
PluginbetaPythonComputer-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 serverbetaPythonWindows automation and durable research MCP server for Codex and Claude Code where Jev chooses among bounded UI Automation actions.
MCP serverbetaCodex computer-use skill where Jev picks the next on-screen element, action, completion and risk from text candidates.
Agent skillprototypeJavaScriptRust MCP server for desktop control via accessibility trees, with an optional decision model (Jev or compatible) for small choices.
MCP serverproductionRust