使用说明
DLARC MCP Framework - AI-Driven Structural Engineering Automation
Research Context
This package is a first pre-step of the research program at DLARC (Deep Learning Architecture Research Center) focused on AI-driven structural engineering automation.
The framework demonstrates how LLM-based AI agents can orchestrate complex engineering software workflows through conversational natural language using the Model Context Protocol (MCP). Rather than requiring engineers to manually operate CAD and analysis software through traditional GUIs, this system enables natural language commands like "Create a 10-floor shear wall building and export it for structural analysis" to be executed automatically across multiple software platforms.
This represents a foundational step toward ASTRA (Autonomous Structural Transformation & Rational Analysis System) - a comprehensive AI structural engineering assistant.
Architecture
+-------------------+ SSH/MCP +-------------------+
| Ubuntu | <-------------> | Windows |
| Ollama + CodeLlama| | AutoCAD 2024 |
| LLM inference | | ETABS v21 |
| Ollama clients | | MCP Servers |
+-------------------+ | pywin32 COM |
+-------------------+
+-------------------+ SSH/MCP ^
| macOS | <-----------------------+
| Claude Desktop |
| MCP clients |
+-------------------+
Platform Roles
| Platform | Role | Key Software |
|---|---|---|
| Windows | Engineering software host | AutoCAD 2024, ETABS v21, MCP servers, pywin32 COM automation |
| Ubuntu | LLM inference host | Ollama, CodeLlama (7B/13B/34B), Ollama MCP clients |
| macOS | User interface | Claude Desktop, Claude Desktop MCP clients |
PLACEHOLDER DATA NOTICE
The following 6 JSON files contain placeholder/example data that users must replace with their own standards data:
| File | Location | Replace With |
|---|---|---|
acixx2_concrete.json |
standards_module/data/materials/ | Your concrete design code (e.g., ACI 318-19) |
acixx_formwork.json |
standards_module/data/construction/ | Your formwork design code (e.g., ACI 347-04) |
asce_placeholder_combinations.json |
standards_module/data/loads/ | Your load combination code (e.g., ASCE 7-22) |
productivity_standards.json |
standards_module/data/construction/ | Regional productivity rates |
ifc4_placeholder_mappings.json |
standards_module/data/codes/ | IFC4 mappings (customizable) |
construction_sequences.json |
standards_module/data/codes/ | Construction sequences (customizable) |
Package Structure
dlarc-mcp-framework/
+-- README.md
+-- example_claude_chat_session.md
+-- windows/
| +-- autocad_mcp_server/
| | +-- autocad_2024_mcp_server.py # Main MCP server (2832 lines)
| | +-- house.py # House generator (all 14 params)
| | +-- autocad_etabs_convert.py # AutoCAD-ETABS converter
| | +-- dxf_to_dataframe.py # DXF file parser
| | +-- construction_validation.py # Standards-based validation
| | +-- shear_wall/
| | | +-- building_dataframe.py # Parametric shear wall generator
| | | +-- building_dataframe_simple.py # Predefined reference building
| | | +-- __init__.py
| | +-- visualization_report_module/
| | | +-- visualization_report_module.py # Report generator
| | | +-- modern_gantt_with_metrics.py # Gantt chart
| | | +-- __init__.py
| | +-- standards_module/
| | | +-- standards_manager.py # Standards query interface
| | | +-- __init__.py
| | | +-- data/ # 6 placeholder JSON files
| | +-- construction_ai_modules/
| | +-- construction_sequencing.py # AI construction sequencer
| | +-- etabs_autocad_pattern_learner.py
| | +-- __init__.py
| +-- etabs_mcp_server/
| | +-- enhanced_etabs_server.py # Main ETABS server (5 tools)
| | +-- corrected_etabs_excel_server.py # Corrected table format
| | +-- fixed_etabs21_server.py # Syntax-fixed version
| | +-- simple_etabs_server.py # Minimal test version
| | +-- minimal_mcp_server.py # Debug server
| | +-- fixed_mcp_integration.py # Ollama + ETABS integration
| | +-- README.md
| | +-- __init__.py
| +-- config/
| | +-- claude_desktop_config.json
| +-- requirements_windows.txt
+-- ubuntu/
| +-- autocad_ollama_client.py # Ollama + AutoCAD MCP
| +-- autocad_ubuntu_client.py # Ubuntu AutoCAD client
| +-- etabs_ollama_client.py # Ollama + ETABS MCP
| +-- unified_ollama_client.py # Unified client (both servers)
| +-- client_integration_ui.py # UI integration module
| +-- ubuntu_install_fix.py # Dependency fix script
| +-- requirements_ubuntu.txt
| +-- ollama_installation_instructions.md
+-- macos/
+-- autocad_mcp_client.py # Claude Desktop AutoCAD client
+-- etabs_mcp_client.py # Claude Desktop ETABS client
+-- claude_desktop_config.json # SSH bridge config
+-- requirements_macos.txt
Quick Start
Windows (MCP Servers)
cd windows/autocad_mcp_server
pip install -r ../requirements_windows.txt
# Start AutoCAD MCP server
python autocad_2024_mcp_server.py
# In another terminal, start ETABS MCP server
cd ../etabs_mcp_server
python enhanced_etabs_server.py
Ubuntu (Ollama Clients)
# Install Ollama (see ollama_installation_instructions.md)
curl -fsSL https://ollama.ai/install.sh | sh
ollama pull codellama:34b
# Install dependencies
cd ubuntu
pip install -r requirements_ubuntu.txt
# Run unified client
python unified_ollama_client.py
macOS (Claude Desktop)
# Copy config (update WINDOWS_IP and USERNAME first)
cp macos/claude_desktop_config.json ~/Library/Application\ Support/Claude/
# Restart Claude Desktop - servers will be available automatically
AutoCAD MCP Server Tools
| Tool | Description |
|---|---|
connect_autocad |
Connect to AutoCAD 2024 via COM |
new_drawing |
Create new drawing |
draw_line / draw_circle |
Basic geometry |
create_building_2d |
2D floor plan |
create_3d_building |
3D building model |
create_house |
Complete house (14 parameters: floors, style, rooms, garage, pool, MEP, etc.) |
create_shear_wall_building |
Parametric or predefined shear wall building |
save_drawing / save_as_dxf |
Save in DWG or DXF format |
zoom_extents |
Zoom to fit all objects |
extract_building_data |
Extract geometry from current model |
generate_comprehensive_construction_report |
Full construction report with Gantt chart |
query_standard |
Query building standards (ASCE, ACI, AISC, IFC4) |
get_load_combinations |
LRFD/ASD load combinations |
validate_for_export |
Check model ready for ETABS/IFC export |
query_aci_318_complete |
Concrete design queries |
query_formwork |
Formwork design queries |
query_productivity |
Construction productivity rates |
Enhanced ETABS MCP Server Tools
| Tool | Description |
|---|---|
test_connection |
Verify ETABS COM connection |
design_building |
Complete structural model (grid, materials, sections, stories, loads) |
export_etabs21_excel |
Generate .xlsx files for ETABS 21 import |
list_exported_files |
List generated Excel files |
generate_import_instructions |
Step-by-step ETABS import guide |
References
- Model Context Protocol (MCP): https://modelcontextprotocol.io
- DLARC Research: Deep Learning Architecture Research Center
- Target journal: Automation in Construction
