MCP-for-anybody

MCP-for-anybody lets an AI agent use AnyBody Modeling System without opening the GUI.

It can:

• create a local AnyBody project copy
• run AnyBodyCon.exe
• extract result objects
• calculate simple summaries such as bilateral knee reaction forces

This repository is a first working version focused on local Windows usage.

Who this is for

If you are new to MCP or AnyBody, start here.

You do not need to understand the full codebase to try it.

What is included

• src/anybody_mcp_server.py: MCP server
• src/anybody_runner.py: local AnyBody automation helpers
• scripts/run_no_gui_workflow.py: no-GUI gait example
• scripts/run_new_posture_workflow.py: no-GUI new-posture example
• start-anybody-mcp.bat: one-click local server launcher
• mcp-config.anybody.json: example MCP client config
• anybody-mcp-workflow.png: workflow diagram

Requirements

This version assumes:

1. Windows
2. AnyBody is installed locally
3. AnyBodyCon.exe exists at:
   • C:\Program Files\AnyBody Technology\AnyBody.7.3\AnyBodyCon.exe

Quick Start

1. Start the MCP server

Double-click:

• start-anybody-mcp.bat

Or run:

D:\Python projects\anybody-automation\start-anybody-mcp.bat

2. Add this server to your MCP client

Use the example config in:

• mcp-config.anybody.json

Its core content is:

{
  "mcpServers": {
    "anybody": {
      "command": "D:\\Python projects\\anybody-automation\\.venv\\Scripts\\python.exe",
      "args": [
        "D:\\Python projects\\anybody-automation\\src\\anybody_mcp_server.py"
      ]
    }
  }
}

3. First test

After your MCP client connects, call:

• get_installation_info

If that works, the server is connected correctly.

Easiest demo

The simplest no-GUI demo is the gait workflow.

Run:

D:\Python projects\anybody-automation\.venv\Scripts\python.exe "D:\Python projects\anybody-automation\scripts\run_no_gui_workflow.py" --project-dir "D:\Python projects\anybody-automation\workspace\auto-gait-run" --overwrite

This will:

1. create a local gait project
2. run it with AnyBodyCon
3. extract left/right knee reaction forces
4. write a JSON report

Expected report:

• workspace\auto-gait-run\Output\knee-force-summary.json

New posture demo

You can also test a new posture without GUI.

Example:

D:\Python projects\anybody-automation\.venv\Scripts\python.exe "D:\Python projects\anybody-automation\scripts\run_new_posture_workflow.py" --project-dir "D:\Python projects\anybody-automation\workspace\auto-posture-run" --preset "forward-lean" --overwrite

Current built-in posture presets:

• forward-lean
• mild-squat
• staggered-stance

Main MCP tools

Useful beginner tools:

• get_installation_info
• create_named_project
• run_model
• get_model_object_value_after_operation
• summarize_bilateral_knee_forces
• run_no_gui_gait_knee_force_workflow
• run_no_gui_new_posture_workflow

Successful GUI-openable example

If you still want a reference project for manual inspection in AnyBody GUI, use:

• workspace\gui-gait-demo\GaitVaughan.main.any

Saved output file:

• workspace\gui-gait-demo\Output\GaitVaughan.main.anydata.h5

Files for advanced users

If you want more implementation details, see:

• DELIVERY.md
• src/anybody_runner.py
• src/anybody_mcp_server.py

Current scope

This first version is best at:

• running existing AnyBody examples without GUI
• creating simple new quasi-static postures
• extracting result objects automatically

It is not yet a full automatic solver for every brand-new complex motion.

Workflow Diagram

See:

• anybody-mcp-workflow.png