How We Prototyped an MCP Server for Geometrid API

The Case for AI in the Built Environment

The architecture, engineering, and construction (AEC) industry continues to grapple with the challenge of making complex building data accessible to diverse stakeholders. In our previous exploration, we demonstrated how Large Language Models could bridge the gap between natural language and construction APIs through a standalone prototype. That work proved the concept, but also revealed an opportunity: what if construction data could be seamlessly integrated into the AI tools that teams already use?

Model Context Protocol (MCP) represents exactly this evolution—a standardized way for AI assistants to connect with external systems and data sources. Rather than building yet another specialized interface, MCP allows construction platforms to integrate directly into established AI ecosystems where users are already working.

Our Geometrid MCP prototype focuses on asset data retrieval—status tracking, progress monitoring, activity feeds, and detailed asset information—all with granular filtering capabilities by date ranges, custom properties, and construction stages. This implementation leverages just a handful of our API endpoints, yet demonstrates the foundation for much more comprehensive integrations.

Breaking Down Barriers with Standardized AI Integration

Model Context Protocol fundamentally changes how AI assistants interact with external systems. Instead of requiring users to learn new interfaces or switch between different tools, MCP enables AI assistants like Claude to directly access and manipulate external data sources through standardized protocols.

For construction technology, this represents a significant shift. Traditional BIM interfaces require specialized knowledge and training. Even our previous LLM prototype, while conversational, still operated as a separate system. MCP eliminates these barriers by bringing construction data directly into the AI assistant that users are already conversing with.

The protocol handles the complex orchestration between natural language understanding and API operations, managing authentication, context, and data flow seamlessly in the background. This creates a unified experience where technical queries about building assets feel as natural as asking about the weather.

Built for Scale with Smart Modular Architecture

Our Geometrid MCP server follows a clean, modular architecture that mirrors the robustness of our API endpoints. At its core, the server acts as an intelligent translator between conversational queries and structured API operations.

• Authentication and Session Management: The server implements secure token-based authentication with automatic refresh capabilities, maintaining persistent sessions across extended conversations. This ensures that users can engage in complex, multi-step workflows without interruption while maintaining enterprise-grade security standards.
• Project Discovery and Resolution: One key insight from our prototype was the need for intelligent project context resolution. The MCP server can automatically select the appropriate project when only one exists, prompt for clarification when multiple projects are available, and maintain project context throughout the conversation.
• Asset Data Retrieval Patterns: The server exposes Geometrid's core data capabilities through five primary functions: authentication, project discovery, asset status tracking, progress monitoring, and detailed asset information retrieval. Each function supports sophisticated filtering by date ranges, custom properties, and construction stages.
• Just-in-Time Data Fetching: Rather than preloading all project data, the server fetches models and detailed information only when needed. This approach optimizes performance while maintaining flexibility, allowing the system to scale efficiently across projects of varying complexity.

Transforming Complex Queries into Simple Conversations

The power of the MCP integration becomes evident in practice. Users can issue queries like "What's the status of precast elements installed last week?" or "Show me progress on curtain wall components for the south elevation," and receive structured, actionable data without navigating complex interfaces.

Consider a project manager asking: "Give me a breakdown of our fabrication progress across all stages." The system would instantly return a comprehensive progress analysis showing that 94% of elements have completed fabrication drawings, 57% are physically fabricated, and 39% have been delivered to site. More importantly, it would highlight critical insights like the installation backlog—nearly 12,000 elements delivered but awaiting installation—enabling immediate action on potential bottlenecks.

The system handles complex parameter extraction automatically, translating natural language descriptions into precise API filters. When a user asks "What's causing our installation delays?" the integration can analyze stage-by-stage progress, identify that only 11% of delivered components are installed, and flag the 6% of untracked elements that need immediate attention. This level of insight would typically require multiple reports and manual analysis, but becomes available through simple conversation.

What's particularly powerful is how the integration maintains context across multi-step workflows. Users can drill down from high-level project overviews to specific asset details, apply filters progressively, and pivot between different data views—all through natural conversation that transforms complex construction analytics into actionable intelligence.

Ready in Minutes with Seamless Setup

Setting up the MCP server requires minimal configuration—essentially just API credentials and a workspace identifier. Once configured, the integration becomes invisible to end users, who interact purely through natural language within their existing Claude interface.

The experience differs markedly from traditional BIM software. Instead of navigating hierarchical menus and learning specialized query syntax, users simply describe what they need. The system handles the technical complexity of API calls, parameter mapping, and data formatting behind the scenes.

This seamless integration extends to error handling and edge cases. When projects aren't found, the system provides helpful guidance. When data isn't available, it explains what's missing and suggests alternatives. The conversational interface feels natural because it handles uncertainty and ambiguity the way humans do.

Transforming Construction Through Conversational AI

The success of our MCP integration points to larger possibilities for the construction technology ecosystem. As AI assistants become increasingly central to professional workflows, the ability to seamlessly access specialized construction data becomes a significant competitive advantage.

MCP enables construction platforms to participate in the broader AI ecosystem without forcing users to abandon their preferred tools. This represents a fundamental shift from requiring specialized software knowledge to enabling domain expertise to flow naturally through conversational interfaces.

The implications extend beyond individual productivity. When building data becomes as accessible as asking a question, it opens possibilities for new types of collaboration, faster decision-making, and reduced barriers to leveraging sophisticated BIM capabilities across project teams.

Why API Quality Determines Your AI Future

Our rapid development of the MCP server was possible because of Geometrid's robust API architecture. Well-designed endpoints with consistent patterns, comprehensive filtering capabilities, and reliable authentication made the integration straightforward rather than complex.

This highlights a crucial point for technical decision-makers: the quality of your underlying APIs directly determines how quickly and effectively you can integrate with emerging AI technologies. Platforms with clean, well-documented APIs can rapidly adapt to new integration opportunities, while those with legacy or poorly designed interfaces face significant barriers.

The construction industry is at an inflection point where AI integration capabilities will increasingly differentiate leading platforms from legacy systems. The ability to seamlessly connect building data with conversational AI represents not just a technical capability, but a strategic advantage in an evolving market.

Expanding into Full Project Management

This prototype demonstrates how a focused implementation—using just our asset endpoints—can deliver immediate value while establishing the foundation for more comprehensive capabilities. This incorporates additional Geometrid endpoints to expand functionality:

• Updating Status: Enable users to change element status based on identifiers, filters, or bulk operations. Users could say "Mark all precast panels on level 3 as delivered" or update entire building systems based on milestone completion, providing flexible status management across project phases.
• Assigning Notes: Enable users to create detailed text entries that add critical context to assets throughout the project lifecycle. Construction teams could dictate inspection findings: "Add note to beam B-205: concrete surface shows minor honeycomb defect, requires cosmetic repair" for quality control and future reference.
• Assigning Labels: Enable users to group and categorize assets for easier management across diverse project needs—defects tracking, procurement items, or custom organizational schemes. Teams could say "Label all HVAC diffusers in conference rooms as 'acoustic priority'" for dynamic grouping and specialized workflow management.
• Assigning Links: Enable users to connect elements with external resources through URLs—tracking links, fabrication drawings, or observation reports. Users could say "Link all steel beams to the fabrication drawings folder," creating seamless information flow between project stakeholders.
• Assigning Dates: Enable users to assign activity deadlines across construction stages to track planned versus actual progress. Project schedulers could say "Set installation deadline for all curtain wall panels to March 15th," enabling real-time schedule variance analysis and proactive delay prevention.
• Assigning Numbers: Enable users to assign numerical values for sequencing, prioritization, or identification—production sequences, installation order, priority rankings, or equipment identifiers. Teams could say "Assign priority 1 to all fire safety components," ensuring coordinated delivery and installation across trades.

Beyond Geometrid's native capabilities, the MCP framework opens possibilities for connecting multiple construction and building management systems. Imagine conversational interfaces that span from design tools to facility management platforms, creating unified workflows that transcend traditional software boundaries.

As we continue to explore the intersection of AI and construction technology, one thing becomes clear: the future belongs to platforms that can bridge specialized domain knowledge with accessible, intelligent interfaces. MCP provides the technical foundation for this future—what matters now is building the robust, flexible APIs that can power it.

Interested in exploring how your construction data could integrate with AI assistants or building your own MCP server for Geometrid API? Contact our team to discuss the possibilities for your specific platform and workflows.