How to connect Claude or ChatGPT to your fleet data with MCP

The Model Context Protocol (MCP) is an open standard that lets an AI assistant such as Claude or ChatGPT call tools in another system. Konvoi runs an MCP server at https://konvoi.ai/api/mcp. Add that URL as a custom connector in Claude, ChatGPT, Claude Code or Cursor, sign in with your own Konvoi login, and the assistant can answer questions from your vehicles, drivers, fines, fuelings and telemetry. It reads with your role and nothing more, and it cannot change anything by itself: every change it suggests becomes a proposal that a person approves in Konvoi Dispatch. AI clients over MCP are in private beta, so Dispatch and AI clients (MCP) must be switched on for your workspace before you can connect.

What MCP is, in plain terms

On its own, an AI assistant only knows what you paste into the chat. MCP gives it a standard way to ask another system for data. That system publishes a list of tools, such as “list vehicles”, “get fuel consumption” or “locate vehicle”, each with a description and the inputs it takes. When you ask a question, the assistant picks the tools it needs, calls them, reads the results and writes its answer.

For a fleet manager, this means you no longer export a spreadsheet to ask an AI about it. You ask “Which trucks were not used last week?” and the assistant gets the answer from Konvoi, with a link back into Konvoi for each vehicle.

For IT: the server speaks Streamable HTTP, stateless, POST only. Sign-in is OAuth with PKCE (S256), and tokens are bound to the server URL. Any MCP client that connects to remote servers over HTTP with OAuth works. The full protocol reference and the live tool list are on the MCP developer page.

What giving an AI access to fleet data means

Connecting a client is a decision about two things: what the assistant can see, and where that data goes.

What it can see. The connection acts as you. It reaches every workspace you can open where Dispatch and AI clients are on, with your role in that workspace. When you allow the connection, you can limit it to some workspaces. If your role cannot see something in Konvoi, the assistant cannot either: a tool whose minimum role is above yours answers with ROLE_REQUIRED.

Where the data goes. Data you ask about is sent to the client you use. Ask Claude about a driver’s fines and those fines go to Anthropic as part of the conversation; with ChatGPT they go to OpenAI. The Konvoi consent screen says so before you click Allow. Check that your company’s AI policy covers the client before you connect it.

The safety model: read with your role, propose the rest

The Konvoi MCP server follows one rule: the assistant reads, people decide.

• Read access with your own role. Read fleet data is always part of a connection, and every call runs with your role in that workspace.
• Changes are proposals. Propose changes is a separate access that you tick, or leave out, when you connect. Even with it, a propose_* tool changes nothing itself. It creates a proposal in Konvoi Dispatch and returns its link.
• Who approves. Everyday proposals, such as assigning a fine to a driver, can be approved or rejected in the client when it can ask you, otherwise in Konvoi Dispatch. Proposals marked Admin approves, such as messaging a driver, requesting a document, merging vehicles or closing a fuel case, need a workspace admin in Konvoi Dispatch.
• Accountability. An approved change runs as the person who approves it, with their role. Every call and every decision is logged.
• Outside text is labelled. Text from outside Konvoi, such as scanned documents and e-mails, is marked as untrusted. Clients are told to treat it as data, never as instructions, so a scanned letter cannot steer the assistant.
• Short-lived tokens. Access tokens last 1 hour. Refresh tokens rotate on every use and lapse after 30 days unused. All your connections are revoked at once when you change your password, log out everywhere or turn two-factor authentication off.

How Konvoi handles your data in general is described on the security page.

Before you start

• A workspace with Dispatch and AI clients (MCP) switched on. This is a private beta. If the client reports DISPATCH_OFF, it is not on for that workspace yet: ask a workspace admin or contact Konvoi to join the beta.
• Your Konvoi login, with your passkey or second step if you use one.
• An MCP client that connects to remote servers over HTTP with OAuth. Claude, ChatGPT, Claude Code, Cursor and VS Code all do. Custom connectors are not included in every plan of every client, so check yours.

The server URL is the same for every client: https://konvoi.ai/api/mcp. Inside Konvoi you also find it under Dispatch → Connect.

Step by step: connect your client

Claude

1. Open Settings → Connectors.
2. Choose Add custom connector.
3. Paste the server URL.
4. Sign in to Konvoi and allow access.

ChatGPT

1. Open Settings → Connectors → Advanced.
2. Turn on Developer mode.
3. Create a connector with the server URL and authentication OAuth.
4. Sign in to Konvoi and allow access.

Claude Code

Run this in a terminal:

claude mcp add --transport http konvoi https://konvoi.ai/api/mcp

Then run /mcp in Claude Code and sign in to Konvoi.

Cursor and VS Code

In Cursor, add the server to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "konvoi": {
      "url": "https://konvoi.ai/api/mcp"
    }
  }
}

In VS Code, run MCP: Add Server → HTTP and paste the server URL. Sign in to Konvoi when asked.

The Konvoi consent screen

Every client ends on the same Konvoi page, “Connect [client] to Konvoi”. It shows who you are signed in as and asks two things:

• Workspaces: All workspaces with Dispatch, or Only these workspaces.
• Access: Read fleet data, which is always on, and Propose changes, which is optional.

Click Allow and you are back in your client, connected.

For scripts and automations without a sign-in window, create a personal token in Profile → AI clients → Create personal token and send it as a bearer token. Personal tokens last 30, 90 or 365 days, and you only see them once.

Example questions to ask

You do not need to know the tool names. Ask the way you would ask a colleague, name the workspace if you have several, and ask for links so you can check the source.

Fleet and vehicles

• “Which trucks did not drive yesterday, and which of them have a tracker that went silent?”
• “Where is 1-ABC-123 right now, and where was it this morning?”
• “What is the total cost of ownership of our five oldest tractors this year?”

Fuel

• “Show September’s fuel consumption per driver, worst first.”
• “Which open fuel cases are there, and what does each one say?”

Fines, documents and permits

• “Which fines came in this week, and who had each vehicle at that time? Propose the assignments.”
• “What expires in the next 30 days: insurance, inspections, permits?”

Konvoi also publishes ready-made prompts, the same playbooks as in Dispatch: Handle this week’s fines, Sweep fuel for anomalies, What expires in the next 30 days and Weekly fleet review. Clients that support MCP prompts show them in their prompt or slash menu.

Limits

The server has fixed limits so one busy connection cannot slow down everyone else:

• Per connection: 120 calls per minute, 5 at a time.
• Heavy tools (analyses and reports that run on Konvoi’s side): 10 calls per minute per connection.
• Per person: 300 calls per minute across all connections.
• Result size: 100 KB. Longer lists are shortened, and the result says what was cut.

Over a limit, the tool answers RATE_LIMITED with the number of seconds to wait.

The practical limits matter more:

• The assistant can be wrong. It can misread a table or draw a conclusion the data does not support. Before you act on a number that matters, open the link and check it in Konvoi.
• It only knows what is in Konvoi. If a fuel card export was never imported, no assistant will find those fuelings.
• It is a beta. Tools and their behaviour can still change. The tool list on the developer page is generated from the server itself, so it is always current.

Removing a connection

• Profile → AI clients lists your connections. Disconnect ends one at once.
• A workspace admin can remove a connection from their workspace in Dispatch → Connect. That person’s connection keeps working for their other workspaces.

The same questions also work inside Konvoi itself, in the Konvoi Dispatch chat, if you would rather not connect an outside client.