Your project data, ready for the AI assistant you already use.
The Visibuild MCP is a secure connection between Visibuild and the AI assistant of your choice – Claude, Copilot, Gemini, ChatGPT, or any other tool that supports the Model Context Protocol. Once it’s switched on for your company, your team can ask questions of their project data in natural language without leaving the assistant they already use.
This page explains what the MCP can do today, where it shines, where it isn’t the right tool, and how we treat your data.
What the MCP unlocks
Three things that weren’t possible before:
Ask questions in plain English. “What’s the health of the Capital Towers project?” or “Show me defects mentioning waterproofing across all our active sites.” The assistant pulls the answer from Visibuild and explains it back to you.
Find issues by meaning, not exact wording. The same defect logged as “water ingress ”, “leak around window head ”, “damp staining bedroom 2 ” and “moisture under sill ” surfaces as one signal. The MCP uses semantic search, so the assistant catches variants that keyword search misses.
Bring Visibuild into your AI workflows. If your team already uses AI assistants to summarise project status, draft reports, or analyse trends, the data they need can now come straight from Visibuild – no copy-paste, no exporting CSVs into a chat window.
What you can do today
These capabilities are live now:
Capability | What it means for your team |
List projects | Pull a list of every project your account has access to, so the assistant knows what portfolio it’s working with. |
List project locations | See the location structure of any project – useful for queries scoped to a specific level, zone, or unit. |
Project health overview | A single-call snapshot of defects, tickets, and recent activity for a project. Great for “how’s this site going? ” questions. |
Semantic search across visis | Search defects, NCRs, inspections, tasks, hold and witness points by intent. Find groups of related issues in one call, across years and sites. |
Semantic search across tickets | The same intent-based search applied to tickets, so issue trends and recurring themes surface even when titles vary. |
Semantic search project templates | Search for templates used across projects and how they related to populated project data including issues. |
Fetch a single visi | Once a relevant defect or inspection is found, the assistant can open it in full – title, status, location, history, attachments. |
Fetch a single ticket | Same as above for tickets – the full record on demand for the assistant to reason over. |
Fetch single project template | Find a template being used on a project. Understand how the template relates to the work being complete. |
Example prompt:
Understand how to improve your templates based on reoccurring issues:
What are my common defects in Visibuild per project and how could we improve our templates to rectify them?
Connecting to the MCP
Connecting takes a minute or two. You'll need three things:
The MCP enabled on your company. MCP is off by default. A user in your company with the
Auth Managerrole can enable it in your company settings:
The MCP URL for your region (see below).
An AI assistant that supports MCP – Claude, Copilot, Gemini, ChatGPT, or any other tool that supports the Model Context Protocol.
Pick the right URL for your region
Visibuild data is stored in regional environments, so each region has its own MCP URL. Use the one that matches where your projects live:
Region | MCP URL |
🇦🇺 Australia | |
🇬🇧 United Kingdom | |
🇺🇸 United States |
Not sure which region you're on? Contact us and we'll confirm it for you before you connect.
Adding it to your AI assistant
The exact steps vary by assistant, but the shape is the same in all of them:
Open your assistant's settings and find the Connectors, Integrations, or MCP servers section.
Add a new MCP server. Give it a name like Visibuild and paste in the URL for your region.
Authorise the connection – you'll be sent to Visibuild to sign in, and access will follow your existing permissions.
Start a new conversation in your assistant and try a question like "What projects do I have access to in Visibuild?" to confirm it's working.
Once connected, the assistant will ask for your permission the first time it uses each Visibuild tool. You're always in control of when it runs.
What makes the Visibuild MCP different
Plenty of platforms now expose their data to AI assistants. Most do it the simple way: wrap every database endpoint as a separate tool and let the AI figure it out. We’ve taken a different approach. Each Visibuild MCP tool is designed around a question your team actually asks, not around a database table.
In practice, that means:
Semantic, not keyword. Searches understand meaning, so the same defect logged five different ways across five teams finally shows up as one trend.
Portfolio-wide reach. Questions like “have we seen this before? ” can run across your whole portfolio, not project by project.
Answers, not data dumps. Each tool is shaped to return a meaningful result in one or two calls, so the assistant spends its time explaining and acting on the answer rather than stitching it together.
Fewer turns, lower cost, less drift. Because the tools are tightly scoped, conversations stay focused and predictable.
A worked example
To show what this looks like in practice, here’s a conversation a QA manager might have with their assistant once the MCP is connected:
You: “Look across waterproofing defects on the last 18 months of projects and tell me what the recurring failure modes are. ”
Assistant: Returns a clustered answer in one or two calls – grouped by location and root cause, drawing on issues logged with different wording across different teams.
You: “Based on that, what should we be checking earlier in the next project? ”
Assistant: Uses the patterns from your real field data to suggest checks targeted at the failures your portfolio keeps hitting.
On a keyword-based interface, the first question alone would take many narrow searches – and would still miss the wording variants. The MCP gets you to the answer in a single conversation.
When the MCP is the right tool – and when it isn’t
Visibuild gives you three ways to get data out of the platform. Each is good at different things. Picking the right one keeps expectations realistic.
Use case | MCP | API | CSV export |
Ask questions of your project data in natural language | ✅ Best | – | – |
Pull project context into an AI workflow (Claude, ChatGPT, etc.) | ✅ Best | ⚠️ Works, heavier lift | – |
Search by meaning across visis and tickets | ✅ Best | ⚠️ Keyword only | – |
Self-serve for non-technical users | ✅ Best | – | ✅ Good |
Bulk data extraction (every defect, every field) | – | ✅ Best | ✅ Good for one-off |
Integrations with Power BI, data warehouses, BI tools | – | ✅ Best | ⚠️ Manual / scheduled |
Static reporting and offline analysis | – | ⚠️ Possible but overkill | ✅ Best |
Real-time / event-driven workflows | – | ✅ Best (webhooks) | – |
Auditable, repeatable queries with exact results | ⚠️ Non-deterministic | ✅ Best | ✅ Best |
✅ Good for
Asking questions in natural language – “what’s the health of Capital Towers? ”, “which defects in this project mention waterproofing? ”
Pulling project context into AI workflows your team already runs
Surfacing patterns in unstructured text – semantic search beats keyword on visi titles and descriptions
Lightweight self-serve once connected – no engineering work needed on your side
❌ Not the right tool for
Bulk data extraction – the MCP returns capped result sets. Use the API or a CSV export instead.
Deterministic reporting where the same query must return identical results every time. The AI layer interprets queries, so framings can vary slightly. Use the API for anything that needs to be exact.
Integrations into other systems – Power BI dashboards, data warehouse syncs, third-party tooling all belong to the API.
Real-time or event-driven workflows – the MCP is pull-based and on-demand. Use API webhooks for triggers.
Compliance and audit trails – if you need a defensible, repeatable export, use CSV or the API, not an AI-mediated query.
Data, privacy and access
The MCP is a new surface for your data, so we want to be precise about how it works.
Off by default. The MCP is not enabled for any company unless we explicitly turn it on with you. You won’t stumble into it - enabling it is a deliberate decision we make together. Reach our to your customer success coordinator to get access.
You stay in control of your data. The MCP is a read interface into the same data you already own in Visibuild. It doesn’t change ownership, sharing, or what’s stored.
Existing Visibuild permissions still apply. Whoever connects the MCP only sees what their Visibuild account can already see. It is not a bypass for project access or roles.
Agents don’t run tools without permission. Your team accepts each request to run a Visibuild tool in their AI interface. Every use is a conscious, opt-in action.
Your choice of assistant, your assistant’s terms. The AI assistant your team connects (Claude, ChatGPT, or another) is your decision, and that provider’s privacy terms govern anything they ask it.
For full details on how Visibuild handles your data, see our Trust Centre.
Getting started
If you’d like to explore using the MCP with your projects, talk to your Visibuild account contact. We’ll walk through the setup, confirm which AI assistant you’d like to connect, and enable it for your company.
If you have ideas for new capabilities you’d like the MCP to support, let us know – the roadmap is shaped by customer requests, and the next set of tools we build will be the ones that unlock the most value for our customers.