🏗️ ClaudeHopper - AI-Powered Construction Document Assistant

What is ClaudeHopper?

ClaudeHopper is a specialized Model Context Protocol (MCP) server that enables Claude and other LLMs to interact directly with construction documents, drawings, and specifications through advanced RAG and hybrid search. It is designed for construction professionals who need to query CAD drawings, plans, and technical specifications using natural language.

How to use ClaudeHopper?

Install Node.js 18+, Ollama (with nomic-embed-text, phi4, and clip models), and optionally Poppler Utils. Run the setup script run_now_preserve.sh for one-click configuration, then place construction drawings in ~/Desktop/PDFdrawings-MCP/InputDocs/Drawings/ and text specs in ~/Desktop/PDFdrawings-MCP/InputDocs/TextDocs/. After adding files, execute ./process_pdfdrawings.sh to ingest documents.

Key features of ClaudeHopper

• Vector-based search optimized for CAD drawings, plans, and specs.
• Visual search to find similar drawings from textual descriptions.
• Specialized metadata extraction for construction industry document formats.
• Efficient token usage through intelligent document chunking.
• Security via local document storage and processing.
• Support for Structural, Civil, Architectural, and other disciplines.

Use cases of ClaudeHopper

• Find architectural drawings for a specific project.
• Locate structural details for a foundation system.
• Search for drawings matching a concrete foundation description.
• Retrieve specifications for interior paint or fire protection systems.
• Identify lift station layout drawings from a document set.

FAQ from ClaudeHopper

What are the dependencies and runtime requirements?

Node.js 18+, Ollama with models nomic-embed-text, phi4, and clip, and optionally Poppler Utils (pdfimages) for image extraction.

Where are my documents stored and processed?

All document storage and processing are local to your machine, using a LanceDB database for vector search. No data is sent to external servers.

What transport or integration does ClaudeHopper use?

ClaudeHopper is an MCP server that integrates directly with the Claude Desktop App via the standard MCP protocol.

How does the image search work?

Images are extracted from PDFs using pdfimages (from Poppler Utils), embedded with a CLIP model, and searched via textual descriptions using the image_search tool.

Can I search across all documents at once?

Yes, the all_chunks_search tool allows searching across the entire document collection, while catalog_search and chunks_search offer more focused queries.