EN / TR
← Back to Products
IN DEVELOPMENT

FraktalMCP

AI for Grasshopper

Connect large language models directly to Grasshopper. Generate components from 387 pre-trained templates, analyze definitions, write scripts, and iterate on designs using natural language. Average response time: 2.3 seconds. Built on the open Model Context Protocol.

# FraktalMCP Example
from fraktalmcp import GHAssistant

assistant = GHAssistant()
assistant.connect()

# Natural language to geometry
assistant.generate(
  "Create a parametric tower"
  " with twisted floors"
)
FraktalMCP - AI-powered Grasshopper interface

Core Capabilities

Natural Language → Geometry

Describe what you want in plain text. FraktalMCP translates to Grasshopper components with 387 pre-trained templates.

Context-Aware

Understands your current definition structure, data trees (list/tree/item), and component connections. Avg. context parsing: 0.4s.

Multiple AI Backends

Works with OpenAI GPT-4, Claude 3, local LLMs (Ollama). Bring your own API key. Response time: 1.8-3.2s depending on complexity.

Definition Analysis

Explains existing definitions in plain English, suggests optimizations (identified 23% redundant components in test definitions), finds null data errors.

Code Generation

Generates C# and Python scripts for GhPython and C# components. Tested on 142 common geometry operations.

Open Protocol

Built on the Model Context Protocol (MCP). Extensible architecture—add custom tools via JSON schema.

FraktalMCP system architecture diagram

Example Prompts

Prompt:

"Create a parametric facade with hexagonal cells that vary in size based on solar exposure"

Result:

Generates attractor-based hexagonal grid with Ladybug solar analysis

Prompt:

"Optimize this structure for minimum material while maintaining stiffness"

Result:

Sets up Karamba analysis with Galapagos optimization loop

Prompt:

"Make the roof panels follow a double-curved surface with planar quad subdivision"

Result:

Creates planarization routine using Kangaroo

Open Source

FraktalMCP is open source and built on the Model Context Protocol. Contributions welcome.