Skip to main content
Glama

Control Interface Generator (CIG)

CIG lets an AI agent operate an ordinary Windows program without screenshots or synthetic mouse clicks.

It works in four steps:

  1. It detects what the program is built with: Win32, WinForms, WPF, WinUI, Qt, Electron or a console program.

  2. It looks that framework up in a capability library, which is data rather than code.

  3. It applies rules that turn each UI element into a typed action, such as set_text, set_checked, choose, menu or answer_dialog.

  4. It serves those actions as an MCP server.

A program is handled through its framework's library entry, never through code written for that one program.

Safety

Every action passes the safety layer before it reaches a program:

  • Reads are always allowed.

  • Writes are refused while the person at the machine is using it, unless the call carries confirm with the person's own words.

  • For third-party programs, a write is refused outright when that program's window is in front and has had recent real input. confirm does not override this.

  • Destructive actions (delete, remove, discard, "Don't Save"... in eleven languages) always need confirm.

  • Password fields, credential and security windows, and browsers are never acted on.

  • Every call is logged as JSON lines, refusals included, with secrets redacted. Files an action is known to change are backed up first, and cig_undo restores them.

  • dry_run checks everything and returns a preview without acting.

Related MCP server: nexusos-semantic

Tools (MCP, stdio)

Tool

What it does

cig_status

Version, loaded adapters, the presence reading and the policy thresholds

cig_apps

Programs with windows that CIG can see

cig_detect

Framework, version and ranked adapters for a program

cig_launch

Starts a program at normal integrity

cig_tree

Compact element tree with stable short ids

cig_actions

The generated typed action set, each with a JSON input schema

cig_do

Performs one action (dry_run, confirm) and reads the result back

cig_read

Reads one element's current state

cig_wait

Waits for an element or window

cig_see

One screenshot of the program's window, on request only

cig_log

The action log

cig_undo

Restores a backup

To register it:

claude mcp add --scope user cig -- py -3 "<repo>\cig_server.py"

Layout

Path

What it holds

cig/

The engine, safety layer, detector and adapters

library/

Framework entries, rules, the destructive-word lexicon and schemas, all data

config/policy.json

Thresholds, plus the allow and deny lists

tests/

Unit and safety tests on recorded fixtures (pytest -m "not live"), and live tests (-m live)

Licence

GPL-3.0-or-later.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables AI agents to seamlessly integrate with the Windows operating system, performing tasks such as file navigation, application control, UI interaction, and QA testing via the MCP protocol.
      -
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI agents to access and act on web, desktop, and mobile interfaces via semantic accessibility trees, using MCP tools for querying, navigating, and executing deterministic UI actions without relying on screenshots or pixel coordinates.
      Apache 2.0
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables an MCP host to observe and control Windows desktop applications deterministically: reading UIA control trees, OCR text and pixel diffs, clicking, typing, scrolling and window management, plus verbatim text reading/appending via ValuePattern/TextPattern. Visual judgment is delegated back to the host model's own multimodal context through annotated screenshots, numbered element cards, grid maps and crosshair zoom-ins, so no separate vision model is needed.
      1
      AGPL 3.0
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI agents to inspect and operate Windows applications through structured UI Automation maps, humanized mouse and keyboard input, and on-demand screenshots or OCR via MCP. It supports locating elements by role, name, or query; clicking, typing, scrolling, window management, event watching, and chained multi-step actions without relying on image analysis.
      MIT