Model Context Protocol (MCP) is an open protocol published by Anthropic in 2024 that standardizes how large language models communicate with external tools, services, and data sources. Before MCP, integrating an AI assistant with design software required custom, one-off API wrappers that broke with every software update. MCP changes this: it defines a universal interface that any AI model and any tool can implement once and use everywhere.
What MCP Does in Practice
MCP servers expose "tools" — named operations that an AI model can call with typed parameters and receive structured responses from. An MCP server for Rhino3D might expose tools like `get_geometry(object_id)`, `create_surface(control_points)`, or `run_grasshopper_definition(file_path, inputs)`. The AI model calls these tools during a conversation or task, integrating design software interaction into the same context window where it is reasoning about the design problem.
For architects and engineers, this means describing a problem in natural language and having the AI operate your design software to explore or implement solutions — without writing custom automation code for every workflow.
MCP Servers for AEC
The AEC sector is building MCP server implementations for the tools that architects and engineers use daily. Current and emerging implementations cover: Rhino and Grasshopper (geometry operations, parametric definition execution), Revit (BIM element queries, family creation, documentation extraction), AutoCAD (drawing operations, layer management), geospatial databases (OpenStreetMap, land registry APIs, terrain data), simulation engines (EnergyPlus, structural FEM solvers), and file formats (IFC parsing, PDF extraction from specifications).
Once a tool has an MCP server, any AI model that supports the protocol can use it — the integration cost drops from weeks of custom engineering to hours of configuration.
Building Your Own MCP Server
MCP servers are lightweight — a minimal implementation is 50–100 lines of Python or TypeScript. The server declares its tools with typed schemas, implements handler functions for each tool, and exposes them via the MCP transport (standard I/O or HTTP). Claude Code, the AI development environment used extensively in MIAW, ships with MCP client support built in — meaning you can connect your own MCP server to Claude Code and immediately use it within any Claude conversation.
For a practice building its first MCP server, the highest-value target is usually the tool they use most: a Grasshopper MCP server that can run parametric definitions from natural language, or a BIM MCP server that answers questions about the active project.
MIAW Module F0 — Foundations
MCP is introduced in MIAW Module F0 as part of the foundational working environment: Claude Code + basic MCP + Git repository. From F0, every subsequent module can optionally leverage MCP connections to the student's design tools. By F8, students build their own MCP servers as part of their adaptive AI ecosystem — connecting the AI infrastructure they have built across all nine modules into a coherent professional practice.