bim2sim-mcp
# BIM2Sim MCP
[](https://github.com/l4b4r4b4b4/bim2sim-mcp/actions/workflows/ci.yml)
[](https://pypi.org/project/bim2sim-mcp/)
[](https://www.python.org/downloads/)
[](https://opensource.org/licenses/MIT)
[](https://ghcr.io/l4b4r4b4b4/bim2sim-mcp)
MCP server for BIM-to-building-energy simulation workflows with IFC extraction, TEASER integration, scenario modeling, weather binding, and results export for downstream WAT/ROI analysis
Built with [FastMCP](https://github.com/jlowin/fastmcp) and [mcp-refcache](https://github.com/l4b4r4b4b4/mcp-refcache) for efficient handling of large data in AI agent tools.
## Features
- ✅ **Reference-Based Caching** - Return references instead of large data, reducing context window usage
- ✅ **Preview Generation** - Automatic previews for large results (sample, truncate, paginate strategies)
- ✅ **Pagination** - Navigate large datasets without loading everything at once
- ✅ **Access Control** - Separate user and agent permissions for sensitive data
- ✅ **Private Computation** - Let agents compute with values they cannot see
- ✅ **Docker Ready** - Production-ready containers with Python slim base image
- ✅ **GitHub Actions** - CI/CD with PyPI publishing and GHCR containers
- ✅ **Langfuse Tracing** - Built-in observability integration
- ✅ **Type-Safe** - Full type hints with Pydantic models
- ✅ **Testing Ready** - pytest with 73% coverage requirement
- ✅ **Pre-commit Hooks** - Ruff formatting and linting
## Quick Start
### Prerequisites
- Python 3.12+
- [uv](https://github.com/astral-sh/uv) (recommended) or pip
### Installation
```bash
# Clone the repository
git clone https://github.com/l4b4r4b4b4/bim2sim-mcp
cd bim2sim-mcp
# Install dependencies
uv sync
# Run the server (stdio mode for Claude Desktop)
uv run bim2sim-mcp
# Run the server (SSE/HTTP mode for deployment)
uv run bim2sim-mcp --transport sse --port 8000
```
### Install from PyPI
```bash
# Run directly with uvx (no install needed)
uvx bim2sim-mcp stdio
# Or install globally
uv tool install bim2sim-mcp
bim2sim-mcp --help
```
### Docker Deployment
```bash
# Pull and run from GHCR
docker pull ghcr.io/l4b4r4b4b4/bim2sim-mcp:latest
docker run -p 8000:8000 ghcr.io/l4b4r4b4b4/bim2sim-mcp:latest
# Or build locally with Docker Compose
docker compose up
# Build images manually
docker compose --profile build build base
docker compose build
```
### Using with Claude Desktop
Add to your `claude_desktop_config.json`:
```json
{
"mcpServers": {
"bim2sim-mcp": {
"command": "uv",
"args": ["run", "bim2sim-mcp"],
"cwd": "/path/to/bim2sim-mcp"
}
}
}
```
### Using with Zed
The project includes `.zed/settings.json` pre-configured for MCP context servers.
## Project Structure
```
bim2sim-mcp/
├── app/ # Application code
│ ├── __init__.py # Version export
│ ├── server.py # Main server with tools
│ ├── tools/ # Tool modules
│ └── __main__.py # CLI entry point
├── tests/ # Test suite
│ ├── conftest.py # Pytest fixtures
│ └── test_server.py # Server tests
├── docker/
│ ├── Dockerfile.base # Python slim base image with dependencies
│ ├── Dockerfile # Production image (extends base)
│ └── Dockerfile.dev # Development with hot reload
├── .github/
│ └── workflows/
│ ├── ci.yml # CI pipeline (lint, test, security)
│ ├── publish.yml # PyPI trusted publisher
│ └── release.yml # Docker build & publish to GHCR
├── .agent/ # AI assistant workspace
│ └── goals/
│ └── 00-Template-Goal/ # Goal tracking template
├── pyproject.toml # Project config
├── docker-compose.yml # Local development & production
├── flake.nix # Nix dev shell
└── .rules # AI assistant guidelines
```
## Development
### Setup
```bash
# Install dependencies
uv sync
# Install pre-commit and pre-push hooks
uv run pre-commit install --install-hooks
uv run pre-commit install --hook-type pre-push
```
### Running Tests
```bash
uv run pytest
uv run pytest --cov # With coverage
```
### Linting and Formatting
```bash
uv run ruff check . --fix
uv run ruff format .
```
### Type Checking
```bash
uv run mypy app/
```
### Docker Development
```bash
# Run development container with hot reload
docker compose --profile dev up
# Build base image (for publishing)
docker compose --profile build build base
# Build all images
docker compose build
```
### Using Nix (Optional)
```bash
nix develop # Enter dev shell with all tools
```
## Configuration
### Environment Variables
| Variable | Description | Default |
|----------|-------------|---------|
| `LANGFUSE_PUBLIC_KEY` | Langfuse public key | - |
| `LANGFUSE_SECRET_KEY` | Langfuse secret key | - |
| `LANGFUSE_HOST` | Langfuse host URL | `https://cloud.langfuse.com` |
### CLI Commands
```bash
uvx bim2sim-mcp --help
Commands:
stdio Start server in stdio mode (for Claude Desktop and local CLI)
sse Start server in SSE mode (Server-Sent Events)
streamable-http Start server in streamable HTTP mode (recommended for remote/Docker)
# Examples:
uvx bim2sim-mcp stdio # Local CLI mode
uvx bim2sim-mcp sse --port 8000 # SSE on port 8000
uvx bim2sim-mcp streamable-http --host 0.0.0.0 # Docker/remote mode
```
## CI/CD Workflow
This project uses a CI-gated workflow to ensure code quality and safe releases:
```
┌─────────────────────────────────────────────────────────────┐
│ Feature Branch → Open PR │
│ ↓ │
│ CI Runs (lint, test, security) │
│ ↓ │
│ ✅ CI Must Pass (enforced by branch protection) │
│ ↓ │
│ Merge to main │
└─────────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────────┐
│ CI Re-runs on main │
│ ↓ │
│ Release Workflow waits for CI Success │
│ ↓ │
│ Docker Images Built & Pushed to GHCR │
└─────────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────────┐
│ Manually Create GitHub Release │
│ ↓ │
│ Publish Workflow verifies Release succeeded │
│ ↓ │
│ Package Published to PyPI │
└─────────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────────┐
│ CD Workflow deploys (staging/production) │
└─────────────────────────────────────────────────────────────┘
```
**Key Safeguards:**
- ✅ Branch protection ensures CI passes before merge
- ✅ Tag pushes verify CI passed before building images
- ✅ Publish workflow verifies Release succeeded before PyPI upload
- ✅ CD workflow only deploys after Release completes
**Manual Gates:**
- 🔒 Creating GitHub Release (allows review before PyPI publish)
- 🔒 Production deployments (requires manual approval)
## Publishing
### PyPI
Configure trusted publisher at [PyPI](https://pypi.org/manage/account/publishing/):
- Project name: `bim2sim-mcp`
- Owner: `l4b4r4b4b4`
- Repository: `bim2sim-mcp`
- Workflow: `publish.yml`
- Environment: `pypi`
### Docker Images
Images are automatically published to GHCR on:
- Push to `main` branch → `latest` tag
- Version tags (`v*.*.*`) → `latest`, `v0.0.1`, `0.0.1`, `0.0` tags
## License
MIT License - see [LICENSE](LICENSE) for details.
## Contributing
See [CONTRIBUTING.md](CONTRIBUTING.md) for development guidelines.
## Related Projects
- [mcp-refcache](https://github.com/l4b4r4b4b4/mcp-refcache) - Reference-based caching for MCP servers
- [FastMCP](https://github.com/jlowin/fastmcp) - High-performance MCP server framework
- [Model Context Protocol](https://modelcontextprotocol.io/) - The underlying protocol specification
TDQS
Scored across 11 tools
Tools are mostly distinct, with clear separation between test context management, cache admin operations, and health check. The only minor overlap is between admin_get_reference_info and get_cached_result, but the distinction (metadata vs. value) is clear.
Non-admin tools follow a verb_noun pattern (enable_test_context, set_test_context, get_cached_result), while admin tools consistently use an 'admin_' prefix. This is a predictable convention, though health_check is a slight deviation.
11 tools is well-scoped for a caching and tracing server, covering test context, cache retrieval, admin functions, and health monitoring without unnecessary bloat.
Cache operations cover retrieval, deletion, listing, info, stats, and namespace clearing, which is comprehensive. Test context has enable, set, reset, and get info. The only gap is no explicit cache creation tool, but that is likely automatic from other operations.