The announcement, published by Revizto on July 28 and republished this week by Construction Management UK, packages three related releases: an MCP Server, a Model & Object Properties API, and a Developer Portal for OAuth 2.0 application registration. The important distinction is that the API and the MCP server are complementary products, not interchangeable access paths. The API is the route for granular model metadata; the MCP server currently centers on coordination data such as issues, clashes, comments, sheets, workflows, reports, projects, and users.
For Microsoft shops, the immediate use case is narrower but still useful. A Copilot Studio maker can add Revizto’s remote MCP server as a tool, authenticate it with OAuth 2.0, and publish an agent that answers questions using the signed-in user’s Revizto permissions. That can reduce the recurring manual work around coordination meeting briefs, unresolved-clash summaries, issue-status reports, and project-risk rollups.
What it does not yet deliver through the MCP connector is a dependable natural-language audit of a model’s object-level properties—such as finding every fire door missing a certification field. Revizto’s marketing material promotes model-auditing scenarios, but its implementation notes list object search, changed elements, object-property reads, and issue-to-object reasoning among features “coming soon” for MCP. Teams planning a Copilot pilot around material quantities, fire ratings, or element metadata should test the API separately rather than assume the connector can retrieve that data today.
Copilot Studio is the supported Microsoft route
Revizto’s setup guide specifically calls out Microsoft 365 Copilot through Copilot Studio, not the consumer Copilot application or a generic Microsoft 365 Copilot chat prompt. In practice, that means this is an agent-building integration. An organization creates or selects a Copilot Studio connection, adds the MCP server through the product’s tool onboarding flow, completes OAuth authentication, and then publishes the resulting agent for other users.
That architecture matters for IT teams. It gives administrators a place to decide who can use the Revizto-capable agent, which connection it relies on, and whether it belongs in a managed environment. It also prevents the misleading impression that every Copilot user automatically gains visibility into every construction project once a tenant enables the connection.
Revizto says the MCP server uses the access rights of the authenticated Revizto user. A user who cannot see a project, license, or account in Revizto should not gain access merely because the question is asked through Copilot Studio. Administrative actions still require administrative Revizto access.
That is the right baseline, but it does not eliminate the normal governance work around an AI agent. A Copilot Studio builder needs to determine whether the agent will act under an individual connection or a shared configuration, who is allowed to publish it, which environments can host it, and whether its outputs might be copied into Teams chats, email, documents, or other Microsoft 365 locations with broader retention and sharing rules.
Revizto’s own instructions reveal a practical wrinkle. Its default MCP configuration uses a public client identifier and no client secret. If Copilot Studio requires a secret for the selected OAuth configuration, Revizto tells administrators to create a custom application in the new Developer Portal, initially with a placeholder redirect URI, then copy the Copilot-generated redirect URI back into Revizto. That is a workable pattern, but it is not a zero-configuration connector.
The real change is an API split, not one universal AI feed
The headline claim—connecting ChatGPT, Claude, Microsoft Copilot, or a private large language model to “live project data”—is directionally accurate. Revizto has made a remote, HTTPS-based MCP endpoint available in its supported hosting regions, including a Virginia endpoint for North American accounts. The server uses Streamable HTTP and OAuth, which makes it compatible with the growing class of AI tools that accept remote MCP services.
But the product boundaries deserve more attention than the announcement gives them. Revizto’s Model & Object Properties API is intended to provide programmatic access to the structured BIM information that makes estimating, carbon tracking, procurement, facilities handover, and compliance checks possible. Those are conventional integration jobs: an application calls an API, retrieves defined fields, and sends the data to a dashboard or line-of-business workflow.
The MCP server presents a different interface. It offers an AI agent a set of controlled tools to inspect and, in some cases, help prepare changes to coordination information. Its current documented strengths are issue and clash analysis, meeting preparation, risk summaries, and task-oriented questions. That is useful for a project manager trying to turn a live issue register into a concise status brief. It is a poor substitute for an API integration that must deterministically retrieve every object matching a property rule.
The gap is especially relevant because the vendor’s examples blur the two. “Which fire doors are missing certification?” requires access to object-level metadata and a reliable interpretation of the relevant property schema. Revizto documents that exact class of information as unavailable through MCP at present. “What are the top five unresolved clashes on Level 3?” is closer to the coordination-data access that its MCP documentation says is available.
For a pilot, project teams should begin with queries whose answer can be verified against an existing Revizto issue or clash view. Ask the agent to summarize open critical issues by trade, produce a coordination stand-up brief, or identify overdue workflows. Do not begin with a safety, code-compliance, cost, or procurement decision that depends on unexposed model properties or on the AI correctly inferring an object relationship.
“Data sovereignty” still requires an AI-provider review
Revizto markets the MCP Server as a data-sovereignty feature and says project data remains in its original region. Its documentation supports one important part of that claim: organizations must use the regional MCP endpoint that corresponds to where their Revizto account is hosted, and a user may connect multiple regions separately. The documentation also says an assistant can combine information from multiple authorized regional connections in a single conversation.
That last detail is the operational caveat. Retaining the project record in Revizto’s regional service is not the same thing as preventing project information from being sent to an external AI tool for inference. Revizto says organizations can bring their own approved AI provider, which leaves the organization—not Revizto—to assess what the chosen Copilot, ChatGPT, Claude, or private-model deployment receives, stores, logs, retains, or uses under its own enterprise agreement.
The vendor’s claim that processing remains “client-side” should therefore be read as a deployment assertion to validate against the selected AI platform’s architecture, not as a blanket assurance that no sensitive information leaves a project system. An MCP session is explicitly designed to pass requests and responses between an AI client and an external data service. The right question for security teams is where the resulting context and output travel after Revizto authorizes access.
Before enabling the service broadly, administrators should document the following controls:
- The organization should restrict the connector to approved Copilot Studio environments and approved Revizto accounts rather than let individual users attach it to unmanaged AI tools.
- The organization should verify that Revizto role assignments reflect current project staffing, because MCP observes the authenticated user’s existing Revizto rights.
- The organization should decide whether prompts and generated responses may include client names, drawing references, design issues, commercial quantities, or safety findings.
- The organization should require human review before any agent-assisted write action, which Revizto itself recommends in its security guidance.
- The organization should test data access in each Revizto region it uses, because separate authorization is required for accounts using different authentication methods.
Windows-based BIM teams gain a new route out of manual exports
The launch is also relevant to Windows engineering environments because Revizto’s publishing plug-ins remain Windows-only. Its current support matrix lists plug-ins for Autodesk Revit, AutoCAD, Civil 3D, Navisworks, Inventor, Advance Steel, Bentley applications, Tekla Structures, and other major design tools; macOS is unsupported for those plug-ins.
That does not mean the MCP connector itself is a Windows-only technology. It is a cloud-facing service that can be consumed by supported AI clients. But many of the teams feeding models into Revizto will be using Windows workstations running Revit, Navisworks, AutoCAD, or Civil 3D. For those teams, the practical promise is less about replacing authoring software and more about shortening the distance between a coordination record and an enterprise reporting or automation workflow.
Revizto has not published separate pricing for the MCP server, the new API, or the Developer Portal. It says the capabilities are available under a Revizto subscription, while its broader integrations material indicates that API access is tied to the higher Revizto+ licensing tier. Organizations should confirm entitlement, API limits, and any Copilot Studio licensing implications before treating the integration as included infrastructure.
The immediate payoff is real if it is kept in scope: a governed Copilot Studio agent can turn the coordination data users are already authorized to see into searchable project summaries and repeatable workflow prompts. The more ambitious promise—AI-driven inspection of BIM-object metadata—will have to wait for Revizto to expose those model-property functions through MCP, or be built now against the separate Model & Object Properties API.
