Between 17 September and 14 October, four freight platforms announced connections for AI assistants. FreightPOP went first on 17 September, according to a Wikifarmer overview of the launches. Flexport followed on 29 September, as FreightWaves reported. Freight Right announced on 7 October, and Cargoplot published its own post on 14 October. All four run a freight MCP server built on the Model Context Protocol (MCP), the open standard Anthropic introduced in November 2024 for connecting assistants to the systems where data lives.
Every announcement promises roughly the same thing: ask your assistant about your freight. What the assistant is allowed to do differs a lot.
What each connector can do
Flexport’s agent can track, quote and book. According to FreightWaves, it can track shipments, surface exceptions or customs holds, search rates, or book outright, and it can return a booking ID and transit time without anyone logging into the platform. Customers can connect Claude, ChatGPT, Microsoft Copilot or their own agents. FreightWaves adds that Flexport says more than 13,000 companies use the platform. Requests through the server are handled by Flexport’s own AI agents, which the company says process 21 million tasks a year. Wikifarmer describes that figure as belonging to the wider AI automation program behind the agents.
Booking has a check built in. Flexport’s product page says price, routing and cargo details are validated before an agent books against a rate it has just pulled. A booking without a rate is held “pending your approval”, and the page does not say who gives that approval, as Wikifarmer points out. Separately, Flexport’s release says every exception routes to a human expert. Where no rate exists, the page adds, the agent requests pricing from Flexport’s team.
Freight Right’s press release covers quoting, tracking and booking across ocean, air and ground. A shipper can hand over a packing list, commercial invoice, purchase order or other shipping document, and the system extracts dimensions, weight, pieces, commodity details and origin and destination to prepare the request. Booking is available today, but the customer confirms the request inside Freight Right before anything is booked. The release also says responses for connected customers can include context from sources such as operations, warehouse activity and customer communications. As an example, it suggests an agent could follow a shipment and surface it if customs clearance has not occurred by a given deadline.
Cargoplot went read-only. Its assistant can look up shipments by reference, container number, vessel, port or city, plus tracking events, documents, invoices and quote requests. In this first version it does not book, change or cancel anything, and it does not download document files. Booking container shipments through the assistant is listed as the next step.
FreightPOP, per Wikifarmer, keeps three write actions (create a shipment, import orders in bulk, cancel a shipment) apart from eight read-only tools. The overview says FreightPOP ties every call to an individual user.
What the fine print says
Flexport’s product page states that any active user can connect once an administrator has enabled the connection for the organization. Each connection belongs to one person signing in with their own account, so the assistant reaches exactly what that person can already reach in the app. Administrators can see every connected assistant and revoke any of them. The MCP server runs separately from Flexport’s public API, with its own rate limits.
Cargoplot’s server shares the public API’s rate limit, and its announcement warns that very long conversations can hit it. The same post says your AI provider can see whatever the assistant retrieves, and advises connecting only assistants your company has approved for shipping data.
Flexport’s product page admits a limit. Asked whether the assistant can analyze historical data such as average transit times by lane, it says “Not yet”. Today’s tools cover live shipments and current rates.
Wikifarmer adds an outside observation: none of the releases it reviewed (Flexport, FreightPOP, Freight Right) names a customer or publishes an error rate for bookings made by agents.
Our reading
We think it is good that logistics providers publish a freight MCP server each. Customers can use the assistant they already have.
Each server, though, shows an agent one provider’s account. Take a shipper who books through two providers. That shipper now has two connectors and two sets of permissions. A question like “which of my boxes arriving in Rotterdam this week are late” could end up with two answers that someone reconciles by hand. That is our assumption. None of the announcements we read addresses the case.
Checks we would run on any vendor, Navo included:
- Which actions can the agent take, and which need a human click?
- Whose permissions does the agent inherit?
- Where does the event data come from, and how old is it?
- Is a rate contract or spot, and on which date was it pulled?
The last one comes from the Wikifarmer overview, which notes that Flexport’s agents search both contract and spot rates.
Where Navo sits
Navo’s homepage describes the problem as status living in twelve tabs: carrier portals that don’t agree with each other, checked by hand one site at a time. The product is one connected data spine across tracking, schedules, rates, loading and air cargo. TrackingMCP covers containers across 229 carriers, normalised to one standard. SchedulesMCP carries 72,000+ sailings on 5,000+ lanes with reliability scoring. FreightRatesMCP publishes daily ocean spot rates with trend analysis. Each product has a dashboard, a REST API, webhooks and an MCP server.
On Navo’s homepage, the products run on shared events, so the lane you priced is the lane you track. The same page says nothing is prepared separately for AI and that the setup has been the same since the first connector. An agent queries the events you read in the dashboard. Event freshness is 0 to 5 minutes, cross-checked across 2 to 3 sources. The user decides how to work: in the dashboard, through the API, or by handing the question to an agent. The homepage does not claim to join accounts held at other providers, so the two-connector case above stays open for us too.
How assistants plug into the same data is described on our AI assistants page. The agent route starts at the Request an Agent page, navo24.com/nav/.
What to watch next
Cargoplot says booking container shipments through the assistant is the next step. Flexport’s product page says backward-looking analysis is on its roadmap, to arrive through a tool it calls Insights Builder. The sources we read give no date for either.
Frequently asked questions
What is a freight MCP server?
It is a connector built on the Model Context Protocol, the open standard Anthropic introduced in November 2024, that lets an AI assistant such as Claude or ChatGPT read data and take actions in a freight platform. What the assistant may do depends on the tools each provider exposes and on the user's own permissions.
Can AI agents book freight through these platforms?
Some can. Flexport's agent can track, quote and book, with validation of price, routing and cargo before booking. Freight Right supports booking but the customer confirms inside its platform first. Cargoplot's first version is read-only, and FreightPOP keeps three write actions separate from eight read-only tools.
Whose permissions does an AI agent use on a freight platform?
On Flexport, each connection belongs to one person signing in with their own account, so the assistant reaches exactly what that person can reach in the app, and administrators can revoke any connection. FreightPOP ties every call to an individual user.
What should I check before connecting an AI assistant to freight data?
Check which actions need a human click, whose permissions the agent inherits, where event data comes from and how old it is, and whether a rate is contract or spot and when it was pulled. Cargoplot also warns that your AI provider can see whatever the assistant retrieves.