MCP control

Pro feature

Model Context Protocol (MCP) control is available in Calib GEOM Pro on the desktop, not in Calib GEOM Web.

MCP lets a compatible AI assistant control the running Calib GEOM window: load camera images, detect features, change calibration settings, optimize, inspect results, export files, and save or open projects. Automation uses the same workflows as the graphical interface.

Enable and disable

1. Start Calib GEOM Pro and sign in to a valid license.
2. Select Help > MCP Control.
3. Configure your assistant as described below.

The menu item is checked while integration is enabled. Your choice is saved and restored when Calib GEOM starts. MCP is disabled by default.

Select Help > MCP Control again to stop the server. Stopping MCP disconnects clients, but does not cancel application work that has already started. Use the application's Cancel control to stop that work.

If startup fails, the log reports the error. The saved enabled preference is retained so the next launch can try again; disable and re-enable MCP to retry without restarting.

Connect a client

Use an MCP client on the same computer that supports stdio. Configure it to launch the Calib GEOM executable with the argument --mcp:

<path-to-calibgeom-executable> --mcp

This starts a headless bridge, not another application window. It connects to the already-running Calib GEOM instance and handles local authentication automatically. No extra Python or Node installation, access token configuration, or connection address is needed.

On macOS, use the executable inside the app bundle, for example /Applications/Calib GEOM Pro.app/Contents/MacOS/Calib GEOM Pro, not the .app directory itself. On Windows, use the installed calibgeom.exe; on Linux, use the installed calibgeom executable.

VS Code example

Add this server to your VS Code MCP configuration, replacing command with the absolute path to your installed executable:

{
  "servers": {
    "calib-geom": {
      "type": "stdio",
      "command": "/absolute/path/to/calibgeom",
      "args": ["--mcp"]
    }
  }
}

Claude Desktop example

Add this entry to Claude Desktop's MCP configuration, using your installed executable path:

{
  "mcpServers": {
    "calib-geom": {
      "command": "/absolute/path/to/calibgeom",
      "args": ["--mcp"]
    }
  }
}

In JSON, Windows path separators must be escaped (C:\\...\\calibgeom.exe), or use forward slashes.

Keep the Calib GEOM window open with MCP enabled. If no enabled instance is running, the bridge reports an error instead of opening a new window. If multiple windows have MCP enabled, disable it in all but the window you want to control; the bridge does not choose an arbitrary project. Restart the client connection after changing the selected window or restarting Calib GEOM.

Only enable MCP for assistants you trust: they can control the application and access files permitted to your user account. Cloud-only connectors that cannot launch a local executable are not supported by this setup.

Available tools

Tools	Purpose
get_status, wait_for_idle	Inspect the operation, recent log, error count, and completion.
set_images, list_images, set_included	Set up camera sources, inspect images, and include or exclude an image or pose.
get_detection_settings, detect_features	Inspect presets and settings, then run detection.
get_optimization_settings, optimize	Inspect camera models and parameters, change settings, and calibrate.
get_results	Read solver and reprojection results, validation, reports, and optionally calibration JSON.
export	Write calibration JSON, an HTML report, statistics, or enabled target-geometry exports.
open_project, save_project	Open or save a calibration project.
cancel	Request cancellation of detection, optimization, project, or export work.

A typical sequence is:

1. Open a project, or set camera image directories or file patterns.
2. Read detection settings, select a preset, and run detection.
3. Call wait_for_idle, then inspect the status and detections.
4. Read optimization settings, change the required fields, and optimize.
5. Call wait_for_idle, inspect results, then export and save a project.

Operations start asynchronously. Starting another operation while work or a setup dialog is active fails. wait_for_idle defaults to 60 seconds and accepts up to 600 seconds; it can return while the application is still busy. Compare errorCount in get_status before and after an operation to detect reported failures, rather than treating an accepted start as successful completion.

Settings patches preserve omitted fields and merge nested objects. Camera arrays take the supplied length, so provide one entry per camera; {} keeps that camera's settings unchanged. Detection automation does not modify saved presets.

Progress, confirmation, and limitations

The server supports the initialized MCP protocol through revision 2025-11-25. The bridge relays progress notifications and confirmation requests. Clients can provide a progress token to receive detection completion and optimizer-iteration updates during wait_for_idle.

Replacing an existing export or project file requires confirmation through the client's form-elicitation interface. If the client does not support that interface, choose a new destination path. Paths must be absolute paths on the computer running Calib GEOM; they are not restricted to the project folder.

Source setup and item-inspection tools are currently camera-oriented. Existing projects can contain LiDAR data and robot poses used by the shared detection and optimization workflows, but dedicated MCP tools for adding those sources and inspecting their individual items are not yet available. Missing project files must be reconnected in the Calib GEOM window. Calibration comparison, metrology import, and diagnostic-map exports are also not exposed as MCP tools.

MCP does not bypass licensing. Tool calls are refused while Calib GEOM is locked. Keep MCP disabled when you are not using automation, and review results in the application before relying on them.