Almost every software product is adding its own AI assistant. That can be convenient - but it is another chatbot chosen by the vendor, with its own limits, pricing, and roadmap, separate from the AI your team already uses.

We decided to build a doorway instead.

ReflectHub now supports MCP (Model Context Protocol), the open standard for connecting AI applications to external tools and data. Compatible assistants such as Claude and ChatGPT can read the ReflectHub project records you are already authorised to see, and answer questions about them in your own words.

Instead of shipping a ReflectHub chatbot, we made ReflectHub work with the assistant your team already knows and trusts. You choose your own AI.

What MCP Means for Construction Teams

It is Thursday afternoon, handover is in eight days, and instead of opening six different views you ask your own assistant:

You ask	Your assistant reads
"What is blocking the handover this week?"	Open tasks, stages, and handover certificates
"How is the Riverside budget tracking against forecast?"	The finance snapshot and expenses
"Which electrical offer is cheapest once alternatives are counted?"	The structured RFQ comparison
"Whose certifications expire in the next 90 days?"	Team and compliance records
"Summarise the last two weeks on site."	The site diary, day by day

Answers can be grounded in the current project record rather than a manually exported snapshot, and where the assistant supports it, it can show which records it used. An assistant can still misread context or make a poor inference - treat its answers as a starting point for judgment. The difference is that it now starts from your records, not from fragments.

What Your AI Can Read

ReflectHub exposes twenty-seven read-only views of the project record. Each one is deliberately curated rather than machine-generated, so an assistant can pick the relevant view reliably.

Readable by your assistant

dashboard

Projects & work

Project overviews, stages, tasks, and comments - who owns what, what is late, and what is blocking the next milestone.

payments

Finance

Budget and forecast snapshots, expenses, and worklog reports - how the project is tracking against the approved budget.

request_quote

Procurement

RFQs, vendor replies, and side-by-side comparisons - including which offer actually wins once alternatives are counted.

engineering

People & compliance

Who is allocated where, which skills you have on hand, and which certifications are about to expire.

history_edu

Site records & handover

Daily site diary entries, current plans, and handover certificates - the day-by-day evidence of the build.

menu_book

Documents & knowledge

Project documents and internal pages, so your assistant reads the current version, not a copy from someone's laptop.

Connecting Takes About a Minute

There are two ways in, depending on where your AI lives:

With a key. Create an access key under Company Settings → MCP - or inside a single project's settings for project-only access - and use it with a compatible MCP client that supports remote connections and key-based access.
With a sign-in. In supported versions of Claude or ChatGPT, add ReflectHub as a custom connector, sign in with your normal account, and approve the requested access. Availability and setup depend on your AI plan and workspace settings.

27

Read-only views

Curated views across projects, finance, procurement, people, site records, and documents.

1

Permission model

Every request is checked against the same access rules as the dashboard.

0

ReflectHub AI add-ons

No proprietary assistant to adopt - use a compatible AI service your team already has.

How ReflectHub Controls MCP Access

Construction data is commercially sensitive, so MCP access follows the same access-control principles as the rest of ReflectHub:

It runs as you. Every request re-runs the permission checks the dashboard enforces, so a connection is limited to the records available to the person who authorised it. If that person's role changes, the connection's access changes with it.
It is read-only. This release ships no write actions. Your AI can read the record; it cannot change it.
It is scoped. A project-scoped connection sees that one project - not your CRM, not your other projects, not your company settings. The scope is enforced on every request.
It is traceable. Every request leaves an audit trail - which view was read, when, and by which connection - without recording the contents of your conversations.
It expires and revokes. Keys carry expiry dates by default, one click in settings ends a connection, and rate limits keep automated traffic in check.

One Project for Your Consultant's AI

Project-scoped access is also useful when working with outside specialists.

Say you bring in an external cost consultant for one development. Hand them a project-scoped key, and their AI can read that project's budget, RFQs, and site diary - and nothing else. When the engagement ends, revoke the key.

Until now, that kind of collaboration meant exports, email attachments, and "please don't forward this". Now it is precise, temporary, and logged.

Why We Chose an Open Standard

Embedded assistants can be convenient, especially for work that stays inside one product. But they also leave the software vendor choosing the model, the interface, the release pace, and the commercial terms. We took a different approach: ReflectHub makes your project data available to the assistant you choose, through an open standard.

Stage

lan

An embedded assistant

hub

Your own AI, connected

Click a row to highlight

MCP separates the system of record from the AI provider. ReflectHub remains the structured record of your construction operations; you choose the assistant that reads it, and can change that choice as the market moves.

The Payoff of Boring Structure

Earlier this month we argued that construction needs structured data before it needs AI. This release is that argument made concrete.

The only reason an assistant can answer "which offer is cheapest once alternatives are counted" is that RFQs in ReflectHub are structured records - vendors, items, replies, comparisons - instead of PDFs buried in an inbox. The only reason it can flag expiring certifications is that people and their credentials are connected data, not rows in a forgotten spreadsheet.

If project information is scattered, connecting an AI gives it faster access to the same confusion. With a structured record underneath, the same connection makes an assistant genuinely useful for project analysis.

Starting With Read-Only

Read-only access is a deliberate first step. Write actions - logging work, drafting a site diary entry - may follow, but only where they can be clearly scoped, explicitly confirmed by the user, and fully audited, under the same permission rules as everything above.

MCP support is available in ReflectHub now, under Company Settings → MCP and in each project's settings. Create a key, connect your assistant, and ask your project a question.

ReflectHub is a project management platform built specifically for construction and architecture firms. It keeps projects, documents, RFQs, costs, and site records connected in one EU-hosted, GDPR-ready workspace - and now, through MCP, readable by the AI you already trust.

Get started → - or book a demo for a guided walkthrough.

Choose Your Own AI: ReflectHub Now Speaks MCP