idfkit-mcp

An MCP server that lets AI assistants work directly with EnergyPlus building energy models. Connect it to Claude, ChatGPT, Cursor, Codex, or any MCP-compatible client and use natural language to:

• Build models from scratch — describe a building and let the agent create zones, surfaces, constructions, schedules, and HVAC systems
• Edit existing models — load an IDF or epJSON file, rename objects, swap materials, adjust setpoints, and validate as you go
• Run simulations — pick a weather file, launch EnergyPlus, and query or export the results without leaving the conversation
• Explore the schema — ask what fields a ZoneHVAC:IdealLoadsAirSystem accepts, what values are valid, and get links to the official EnergyPlus documentation
• Search the docs — full-text search across the EnergyPlus I/O Reference, Engineering Reference, and other documentation sets hosted on docs.idfkit.com

Built on idfkit, it supports EnergyPlus 8.9 through 26.1 (17 versions with bundled schemas). Schema exploration, model editing, and validation work out of the box with no external dependencies. Running simulations requires a local EnergyPlus install — the server discovers it automatically via PATH, the ENERGYPLUS_DIR env var, or standard OS install locations. A Docker image with EnergyPlus bundled is also available.

[Documentation](https://mcp.idfkit.com/docs/) | [GitHub](https://github.com/idfkit/idfkit-mcp/)

Tools

The server exposes 37 tools across ten categories, plus 13 MCP resources for read-only data access:

All tools return structured Pydantic models. Schema, validation, and search results include direct doc_url links to the relevant EnergyPlus documentation.

Session state (loaded model, simulation results, weather file) is persisted to disk automatically, so clients that restart the server between turns (e.g. Codex) can resume where they left off.

Installation

pip install idfkit-mcp

Or with uv:

uv add idfkit-mcp

Usage

Run as stdio MCP server (default):

idfkit-mcp

Run as Streamable HTTP MCP server:

idfkit-mcp --transport streamable-http --host 127.0.0.1 --port 8000

Quick MCP Setup

Add idfkit-mcp to your MCP client. Example for Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "idfkit": {
      "command": "uvx",
      "args": ["--from", "idfkit-mcp", "idfkit-mcp"]
    }
  }
}

See MCP Client Setup for all supported clients (Claude Desktop, Cursor, VS Code, Claude Code, Windsurf, ChatGPT, Codex, JetBrains, Cline, Continue, and Zed).

Development

This project uses uv for dependency management and Zensical for documentation.

Setup

# Clone the repository
git clone https://github.com/idfkit/idfkit-mcp.git
cd idfkit-mcp

# Install dependencies and pre-commit hooks
make install

Commands

make install    # Install dependencies and pre-commit hooks
make check      # Run linting, formatting, and type checks
make test       # Run tests with coverage
make docs       # Serve documentation locally
make docs-test  # Test documentation build
make docker-build  # Build base Docker image (no EnergyPlus)
make docker-build-sim ENERGYPLUS_TARBALL_URL=<linux-tarball-url>  # Build simulation image
make docker-build-sim DOCKER_PLATFORM=linux/amd64 ENERGYPLUS_TARBALL_URL=<linux-x86_64-tarball-url>  # Apple Silicon + x86 tarball
make docker-run    # Run Docker container

Releasing

1. Bump the version: uv version --bump <major|minor|patch>
2. Commit and push
3. Create a new release on GitHub with a tag matching the version (e.g., 1.0.0)

The GitHub Action will automatically publish to PyPI.

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.