Visibuild MCP Connecting Construction Project Data to your AI Assistant.

What is MCP?

Model Context Protocol is an open standard, developed by Anthropic and now widely adopted, that lets AI assistants connect directly to external software. Instead of the user acting as the bridge between their AI tool and their platform data, the platform provides the connection itself.

In practical terms: connect Visibuild to your AI assistant, and it can query your project data in real time, in response to questions you ask in plain language. No export, no copy-paste, no manual translation.

What the Visibuild MCP lets you do today

Nine tools are live today, each designed around a real question a site team would actually ask:

search_visis semantic search across defects, NCRs, inspections, tasks, and hold/witness points. Ask about “waterproofing issues” and get back everything relevant, even if it was logged as “damp staining,” “moisture under sill,” or “leak around window head.”

search_tickets the same semantic capability across your tickets.

search_project_template search across all project templates.

get_project_template look into specific project templates

get_project_health defect, ticket, and activity overview for a project in a single call.

list_projects / list_project_locations navigate your portfolio structure.

get_visi / get_ticket / get_project_template fetch individual records by UUID.

The MCP is available at regional endpoints: AU, EU, and US. It’s opt-in and off by default, no company is connected unless it’s a deliberate decision made together with that company.

Why semantic search changes everything

Construction data is almost always richer than keyword search can see.

The same defect gets logged differently depending on who’s on site, what their background is, and what tool they used. “Water ingress at level 7 slab” and “damp staining bedroom 2 wall” and “moisture below window head” might all be the same problem, but historically, there was no way to connect those records unless they happened to be described identically. Related data points stayed unrelated, invisible to any analysis that relied on consistent language.

The Visibuild MCP tools are designed to answer your questions as fast as possible without your AI tools first needing to process every single piece of data before it can reach a conclusion. This is made possible through Visibuild MCP’s RAG (Retrieval-Augmented Generation) search tools. These tools allow for board search terms that find all related data instantly and across all of your projects, helping you quickly identify trends and uncover hidden patterns.

This matters most for portfolio-wide analysis. A question like “what waterproofing failures have we seen in the last 18 months across all our projects?” is only useful if the search actually finds all of them, not just the ones your team happened to label consistently.

Why fewer tools is the whole point

Every Visibuild feature starts with the same question: what does someone in construction actually need to know? Not what’s technically possible to surface. Not what would make for an impressive list of capabilities. What would genuinely help a site team, a quality manager, or a project director do their job better.

The Visibuild MCP is no different. Instead of asking “what can we expose” we asked “what would a site team actually want answered?”, the questions they’ve been dying to ask their data for years but couldn’t. So, we built the tools around those questions.

Internally, we’ve using ‘The Pantry and The Chef’ analogy to explain this. A great chef doesn’t need every ingredient in the kitchen, they need the right ones, prepared well, and ready to go. The quality of the meal doesn’t come from the size of the pantry, but rather comes from knowing exactly what to cook, and having exactly what you need to cook.

Some platforms have approached MCP by taking every REST API endpoint they have and wrapping each one as a separate tool. The result looks comprehensive, dozens or even hundreds of tools.

In practice, we believe it’s the opposite of useful.

When an AI assistant has 50 or 100 tools to choose from, it has to figure out which ones to call, in what order, to answer a single question. More chains of calls. More chances to get it wrong. Slower, more expensive answers. And because these tools are just raw endpoints, the search is still keyword-only. The AI can’t compensate for meaning mismatches by trying harder.

What it looks like in practice

The best illustration is the waterproofing analysis loop. Here’s a real workflow:

1. “Look across waterproofing defects on the last 18 months of projects and tell me the recurring failure modes.”
2. The AI calls search_visis, gets a semantically clustered result set across your portfolio, and returns a grouped answer, by location and root cause.
3. “What would you change in our waterproofing inspection template to catch these earlier?”
4. The AI fetches the active template and proposes targeted additions.

Steps 1 and 2 are possible on an endpoint-wrapped MCP, but only for defects literally labelled “waterproofing.” Steps 3 and 4 aren’t possible at all.

Here’s another. A portfolio manager asks:

1. “Skycity Hospital project has 20 major waterproofing defects. Capital Towers has none. What did we do differently?”
2. The Visibuild MCP pulls inspection records, template versions, hold point completion rates, and trade sign-off patterns across both projects, and surfaces the differences.

Same question type. Same single conversation. The kind of cross-project comparison that previously required hours of manual data work, if it was possible at all.

What the MCP is not for

MCP is a conversational interface, and it’s important to be clear about its limits. For bulk data extraction, every defect across every project, complete field sets, the Core API is the right tool. For Power BI dashboards or data warehouse syncs, the API with webhooks. For compliance-grade, reproducible audit exports, CSV or API every time.

The MCP is for asking questions and getting answers. For single conversations that pull project intelligence without requiring engineering work on the customer side.

What’s coming next

The current toolset covers the core question types: project health, defect and ticket search, individual record retrieval. The roadmap includes template read/write, milestones, documents, attachments, and proactive risk surfacing.

Every addition will be justified the same way: is there a real question from a real site team that this answers better?

Getting access

The Visibuild MCP is available to customers on request. It’s off by default, reach out to your CS contact or [link: request access] to get set up.

The Visibuild MCP is available to all customers. Your company’s authorisation manager can enable it directly: head to Company Settings, toggle MCP on, and your unique MCP URL becomes available immediately, no request to Visibuild required.

Set up guides for Claude, Copilot, ChatGPT and other compatible assistants can be found here.