ProjectTeam, Inc. announced on 4 November, from a Chantilly, Virginia dateline, that its construction management platform now connects to commercial AI assistants over the Model Context Protocol. The release went out via EIN Presswire and leads on a line from Ty Witmer, its president and founder: “Feed AI fragmented records and you get fragmented answers.”
That framing — connected data as the precondition for useful AI — is the company’s existing argument about its CoreLink Data Model, announced in June. The newer and more specific material is not in the release at all. It is on the company’s own AI product page, which enumerates what the connection can and cannot do more precisely than most of the MCP documentation this publication has read.
What Was Announced
ProjectTeam.com connects through an MCP server. The product page names four assistants — “ProjectTeam.com connects to Claude, ChatGPT, Copilot and Gemini so your team can use AI with real project data” — and describes setup as adding the MCP server to the AI tool and signing in with existing ProjectTeam credentials.
On the read side, the AI can reach “projects, documents, schedules, costs, action items, and contacts,” and the page is explicit about retrieval scope: “The AI pulls only the records needed to answer each request, based on what your account can access.”
On the write side the page gives a list — introduced with “including,” so not stated as exhaustive. The connected tool can create or update records on supported forms, “including: RFIs, Punch List Items, Field Reports, Directives, Contract Exhibits, and Custom Forms.” It can also share documents with the right share groups and help administrators set up form types and fields. The release adds photo and voice capture — turning notes and spoken updates into structured records — and “dashboards built from live project data inside the AI assistant, refreshed on demand.”
Availability and price are both stated. The release says the connection “is available today to ProjectTeam.com Subscribers in the platform’s commercial environment and is included in each subscription”; the product page says “AI access is included for ProjectTeam.com Subscribers as part of the current platform price.” Access is restricted to Subscribers, meaning Standard Users and Organization Administrators, with Contributors able to upgrade. Asked whether a paid assistant plan is needed, the page says that for ChatGPT it is: the connection uses Developer Mode, “available on Plus, Pro, Business, Enterprise, and Edu plans” — those being ChatGPT’s plans, not ProjectTeam’s. Claude connects on desktop and web, and on Claude Team and Enterprise plans an organisation owner may need to add the connector first.
Note the scoping in the release’s own availability sentence: it places the connection in the platform’s commercial environment. The release does not extend that statement to the federal environment, and this publication is not going to infer either way about it.
The Permissions Line Is Table Stakes Now. The Ceiling Above It Is Not.
ProjectTeam’s security list reads the way the market now expects: “Role-based and field-level permissions on every request,” single sign-on support, encrypted transfer between the platform and the AI tool, “AI access only for organizations that choose to turn it on,” and a user-level disconnect at any time. The audit answer is there too: “Records created through AI appear in the platform with the same activity history as every other record,” including who changed what and when.
This publication should say clearly that none of that is novel, because it has made the argument itself more than once. Reporting on Access Evo in October, we wrote that “inheriting the host system’s permission model rather than building a second one is the correct answer, and it is the specific risk that an AI layer over an ERP introduces.” CMiC reached the same conclusion and put it in a press release at the start of this month; Qonic reached it from inside a BIM model days later; Revizto, Access Evo and Graneet are all on the same page. Permission inheritance is a settled answer, not a feature.
Less commonly spelled out in a single sentence, in the documentation this publication has read, is what comes next: “Some actions stay in ProjectTeam.com, including deleting records, editing schedule and cost forms, creating projects, and managing users.”
Analysis. That is a different kind of control from permission inheritance, and the distinction is worth drawing because the two are easy to conflate. Permission inheritance says the agent can do what you can do. A capability ceiling names things the connection does not do. The four named here fall into two groups. Three are destructive or structural — deleting records, creating projects, managing users — the actions whose blast radius is the project or the tenancy rather than a document. The fourth is editing the schedule and cost forms.
That fourth one is the interesting one on a capital-projects platform, because schedule and cost are where the money and the claims live. The AI can create a contract exhibit and a directive — instruments with contractual weight — but not edit a cost form. The page does not explain the reasoning, and this publication is not going to supply one on the company’s behalf. It is simply a clearer statement of scope than this publication usually finds, and a reader evaluating the product can act on it.
The nearest comparison in our coverage is the Autodesk pattern, where the Write and Experimental Revit MCP servers arrived as installers separate from the original read-only one and the capability is chosen by which server you connect. ProjectTeam’s is one connector with a fixed surface. Build Paperless took a third route, scoping by issued key. Three shapes of the same decision, and no shared vocabulary yet for comparing them.
The Customer Base Is the Reason to Pay Attention
ProjectTeam sells into capital programmes, and its own material names owners, contractors, subcontractors and public agencies as users — a different seat from most of the MCP connectors this publication has covered. The company’s own FedRAMP page describes the platform as “FedRAMP Authorized and Certified” and “Listed on the FedRAMP Marketplace,” says it is GovRAMP Authorized for state and local government cloud security standards, and names the National Park Service, the DC Department of General Services and the Maryland Stadium Authority among its users. Those are the company’s own characterisations and customer claims, reported here as such; this publication could not independently query the FedRAMP Marketplace, whose listing interface did not return machine-readable results.
Most of the MCP announcements this publication has tracked this autumn sit with design tools, model authoring, field capture and contractor ERPs. An owner-side capital-programme record is a different object. The questions a public agency’s project controls team asks — where is this change order, what did we approve, who signed it — are the shape of question this kind of connector is built for. They are also the questions a public records request asks, which is why the activity-history sentence matters more here than it would on a design tool.
What the documentation does not establish, and should not be read as establishing, is how any of this performs. It gives no accuracy figure, no benchmark, no agency named as using the connector, and no statement about what happens when the model misreads a record it has permission to update. The company has published a clear scope. Scope is not evidence.
