Skip to content

SmartHopper 2.0.0-dev.260705: MCP Server and Canvas Automation

Pre-release
Pre-release

Choose a tag to compare

View all tags
@github-actions github-actions released this 05 Jul 20:38
· 29 commits to main since this release
2.0.0-dev.260705
24f3ea4

This dev release brings a built-in MCP server so external clients can drive SmartHopper tools, expands the canvas automation toolkit with five new tools, introduces component protection for mutating AI operations, and polishes PDF/web-to-Markdown conversion with image-mode controls.

🌐 MCP Server

New SmartHopperMcpServerComponent starts a local HTTP/JSON-RPC MCP server that exposes SmartHopper tools to external clients such as Claude Desktop or other MCP-compatible apps. The component is opt-in, restarts automatically when port, bearer token, or mutating-tools inputs change.

🧰 Canvas Tools

Five new tools for automating the Grasshopper canvas:

  • gh_connect and gh_disconnect to wire or unwire components
  • gh_remove to delete components
  • gh_document_save to save the document
  • button_click to trigger button components and force downstream recalculation

gh_put and gh_get now paginate and support Count/pageSize limits, and gh_put reliably restores external connections.

🛡️ Canvas Protection

Components that implement ICanvasProtectedComponent (starting with the MCP server component) are protected from mutating AI tools, so critical components are not accidentally modified by automated canvas operations.

📝 Markdown & File Conversion

PDF-to-Markdown conversion now preserves hyperlinks, detects lists, interleaves inline images in reading order, and cleans up numbering. Web-to-Markdown supports link, embed, describe, and caption image modes, with fixed edge cases for Wikipedia and Discourse. A new AIWeb2MdComponent provides AI-powered URL-to-Markdown conversion while the standard Web2MdComponent is now non-AI.

📊 Live Runtime Data

Canvas query responses can now include live Grasshopper parameter values as runtime data via the shared GhJSON library, so external tools can inspect what the definition is actually computing.

🔍 Self-Documenting Tools

New smarthopper_workflows and smarthopper_tool_help tools help MCP clients discover available workflows and tool usage. Every AI tool now exposes MCP metadata, JSON schemas, and read-only/mutating hints.

⚙️ Provider Settings & Models

Provider settings were refreshed: Anthropic adds ServiceTier and ReasoningEffort; MistralAI and Gemini add ReasoningEffort; Gemini restricts SafetyLevel to allowed values. Model registries were updated across Anthropic, DeepSeek, Gemini, MistralAI, OpenAI, and OpenRouter.

🛠️ Technical Requirements

  • Rhino 8.0 or above is required
  • Windows 10/11 or macOS
  • Valid API keys for MistralAI, OpenAI, DeepSeek, Anthropic, Gemini or OpenRouter

⚠️ Important Notes

  • This is a dev release with some features still unstable or subject to change
  • API keys are required, and usage costs apply based on your provider
  • Documentation is currently under development

🤝 We Value Your Feedback!

Help shape SmartHopper's future by:

  • Sharing your experiences with the new features
  • Suggesting improvements via our discussion
  • Telling us what AI capabilities would help your workflow most

We hope you enjoy these new features and improvements!

Happy designing! 🎨

Assets 5
Loading