ChatGPT integration: ask Roadmark from ChatGPT

ChatGPT connects to outside systems through apps, and a custom app can be built on any remote server that speaks the Model Context Protocol (MCP). Roadmark's connector is an MCP server, so the Roadmark app in ChatGPT is one your workspace admin creates from its address, tests and publishes. There's no code to write.

What your team can ask ChatGPT

Anything they could find in Roadmark, asked in their own words: which loads picking up tomorrow still have no truck, which carrier bills don't match their rate con, how much detention was billed in September. See the AI assistants page for more questions by team.

For example, with Roadmark's example group, someone in sales selects Roadmark in a message and asks why the last five Northgate Home quotes were lost. ChatGPT reads those quotes in Roadmark, with the lanes, the prices and the outcome recorded on each, and answers in the chat, so a pattern, such as the same lane lost on price each time, shows up without anyone building a report. If the rep then asks for a new price on that lane, the Quoting agent prices it from the same contracts and history it always uses, and the quote waits in Roadmark as a draft for approval. Northgate Home is set to act with approval, so nothing goes to the customer until a person sends it.

What ChatGPT needs

OpenAI's help article on developer mode and MCP apps sets out who can do what:

• Plans. Apps with full MCP support, including write actions, are for ChatGPT Business, Enterprise and Edu, on the web. OpenAI describes full MCP support as rolling out in beta. Pro users can connect MCP servers in developer mode with read and fetch permissions only.
• Who sets it up. Only admins and owners can turn on developer mode on Business, or publish an app. On Enterprise and Edu, admins can also give developer mode to named people and choose, with role-based access, who may use each app.
• Mobile. Custom MCP apps are web only for now.

Two sets of permissions, both enforced

Roadmark's. The app runs as the person asking and carries exactly their access. A Roadmark admin sets how far assistants may go, from read only, to read and draft, to acting within the rules your team has set for the agents. Work asked for in ChatGPT goes through the same agents as work asked for in the app, so a dollar limit on a lane, a customer on credit hold or a carrier blocked by compliance applies the same way. See AI controls.

ChatGPT's. OpenAI's admin controls let a workspace decide when ChatGPT asks before using an app, with options such as always ask, allow read actions, or allow low-risk actions. On Enterprise and Edu, admins can also switch individual read and write actions on or off. ChatGPT asks for confirmation before write actions based on those permissions and the action's context.

Every action taken through ChatGPT is on the record in Roadmark, with the person, the assistant and the time.

When Roadmark's tools change

After an admin publishes an app, ChatGPT keeps a fixed snapshot of its tools. When Roadmark adds or changes a tool, the workspace doesn't pick it up on its own: on Enterprise and Edu, an admin refreshes the app's actions from Workspace settings, where new actions start switched off. On Business, OpenAI says a published app can't be updated yet, so an admin recreates and republishes it.

Using it day to day

In a chat, select Roadmark for a message, or @mention it. The selection applies to that message only, so mention Roadmark again when a follow-up needs fresh data. When a request needs work done, the result lands in Roadmark as a draft, with the reasons, in the same queue your team approves everything else from.

The same connector works in Claude and Gemini too, and the Developers page describes the MCP server behind it for teams building their own tools.