MCP Connector for Navisworks

What it does

A Navisworks add-in that lets your existing AI assistant read and interrogate the model that is open on the user's machine, including clash results, and in the Pro edition author review content such as clash tests, saved viewpoints and selection sets, open, append and save model files, and write export files. You bring your own AI subscription. Supported on Navisworks Manage and Simulate 2022 to 2027. Clash tools need Manage.

How edits are controlled

The connector carries out calls from programs running under the same Windows account, normally your AI client. Whether you approve each call is set in the AI client. The connector does not require a separate confirmation for every edit.

The free tier is read-only. Its tools interrogate the model, its properties and its clash results, and can move the camera, but cannot change model content, write files or save.

Pro tools can author review content in the open file: selection and search sets, saved viewpoints, comments, clash tests and results, TimeLiner tasks and appearance overrides. Those changes run inside named Navisworks transactions, so each appears as a named step in the Navisworks undo history. Navisworks cannot roll a transaction back automatically, so a tool that fails part-way can leave a partial change, and batch tools keep the items that succeeded. Appending a model, opening or starting a document, the background colour, section planes, the selection and the camera are not recorded as undo steps.

No tool is designed to write back to the source models appended to the scene. Exports go to the path named in the call and overwrite a file already there, within the user's Windows permissions. The connector never saves on its own: the open file is written only when the AI calls the save tool or names that file as an export path, and neither asks for confirmation in Navisworks. Opening or starting a new document is refused while a model is open unless the call sets an explicit flag to discard unsaved changes.

Where model data goes

Tool results go to the AI application you connect. If it uses a cloud AI provider, project information is processed under your account settings and agreement with that provider. A local AI setup can keep inference on your own computer or server. Running the connector locally does not, by itself, make the AI processing local.

The add-in runs inside Navisworks on the workstation. The component that talks to your AI client makes no outbound network connections of its own.

Project files and outputs

The connector does not send model content or tool results to AUTOM8LABS. Your AI application controls what it sends to its provider. Files, images and reports requested through tools can contain project information and can be written to locations available to your Windows account, including network shares.

What AUTOM8LABS receives

The machine identifier is a SHA-256 hash of hardware identifiers, or of the machine and account names where Windows cannot supply them. The underlying values never leave the machine.

What is stored

On the workstation

• No listening port. The add-in communicates with your AI client over a Windows named pipe, not a network socket.
• Access restricted to the same Windows user. The pipe is scoped to the account Navisworks is running under, and no other account on the machine can open it. Any program running under that account can use it, and the pipe is open whenever Navisworks runs with the add-in. A pipe that cannot be scoped that way is never created.
• No log file by default. Diagnostics are off unless a user sets an environment variable for troubleshooting. When enabled, the log records start-up, licence and AI app setup events, with licence keys masked, and does not record tool calls or model content. Files are kept for 30 days and never leave the machine.
• Stored credentials are encrypted. The licence cache is protected with Windows DPAPI at machine scope, so a copy taken to another computer cannot be read. A key placed by IT for pre-deployment stays unencrypted until it is first validated.
• Adds itself to AI apps. Setup adds the connector to the MCP settings of AI apps it finds on the machine. Switching AI apps on in the connector dialog adds it to every supported app, including Codex inside WSL, which it reaches by running wsl.exe. Switching off removes it. Each time Navisworks starts, the add-in repairs its own entry if the connector has moved, and changes nothing else in those files.
• No elevated privileges are required at runtime. Installation needs administrator rights once.

Administrator controls

With Pro, an administrator can block tools by name or permit only an approved list. These restrictions apply when tools are listed and when a call is executed, including calls inside a batch. They restrict the licensed tools; they do not grant additional access. The controls are off unless configured.

Policies can be assigned by computer or by user through Group Policy or Intune using the supplied template. Where policies overlap, the stricter tool restrictions apply. An administrator can also force Read-Only Session; the user cannot turn that lock off in the connection dialog.

Read-Only Session restricts the connector to its free inspection and navigation tools, even when a Pro licence is active. It is enforced inside the host, as well as in the list shown to the AI.

Optional audit log

The optional Pro audit log is off by default. When an administrator enables it, the connector records tool calls, including refusals and batch steps. Records contain the Windows user and computer, host version, project name or path, AI app name, tool, outcome, duration and shortened arguments and error text. These fields can contain project information. Tool results are not copied into the audit log.

An administrator can also request a short reason from the AI for each call. This is the AI's account of the request, not a copy of the user's chat. Logs are written locally and can be copied to an administrator-selected folder or network share. Local audit files are retained for 90 days; retention and access at the destination are controlled by your organisation. AUTOM8LABS does not receive them.

Audit recording is best effort. A failed write or unavailable share does not block the tool, so the log is not a guaranteed or tamper-proof record.

Supply chain

The add-in ships a single third-party .NET component, a widely used JSON library from the official NuGet ecosystem, under the MIT licence.

The component that talks to your AI client is a self-contained executable built on a pinned Node.js runtime. Its own dependency set is MIT, ISC or BSD licensed. There are no copyleft licences in the shipped product.

Dependency inventories cover the plug-in and its AI connection component, including the bundled runtime. CycloneDX SBOMs are available on request for this release. Release package checks on 12 October 2026 reported no known vulnerabilities in the dependencies checked. This is a dated check, not a guarantee against vulnerabilities discovered later.

SBOMs are available on request from [email protected].

Assurance

Every AUTOM8LABS binary and the installer are code-signed as AUTOM8LABS LTD through Azure Trusted Signing. The JSON library ships unmodified with its publisher's own signature.

See the Trust Center for the full assurance position and the sub-processor list.