CargoAi has unlocked its CargoMART air cargo marketplace for AI assistants, including Microsoft Copilot, using the open Model Context Protocol. Starting June 5, 2026, freight forwarders and shippers can search, book, and track air cargo shipments directly inside Copilot, ChatGPT, and Claude—turning a chat window into a full-fledged cargo desk.

CargoMART is an online marketplace where freight forwarders compare air cargo rates and capacity from hundreds of airlines and GSAs. Until now, users interacted with it through a web browser or API. By building an MCP server, CargoAi hands the marketplace to AI agents that can act on behalf of logistics professionals. The server exposes CargoMART’s search, booking, and tracking tools, so any MCP-compatible client—Copilot included—can call them during a conversation.

For Windows enterprise users, this matters because Microsoft is betting heavily on Copilot as the fabric of productivity, and MCP is the wiring that connects Copilot to real-world business tools. Instead of switching between a browser, a TMS, and email, a logistics coordinator can ask Copilot, “Find me the cheapest next-flight-out option for 2,000 kg of pharma from Frankfurt to Chicago,” and get live results pulled from CargoMART in seconds.

How MCP turns Copilot into a transaction engine

Model Context Protocol (MCP) is an open standard that lets large language models discover and use external tools securely. Think of it as USB-C for AI: one connector, many devices. An MCP server describes its capabilities in a machine-readable manifest—available tools, resources, and prompts—and a client like Copilot can then invoke those tools on the user’s behalf.

When CargoAi launched its MCP server for CargoMART, it essentially published a set of function signatures: search_flights(origin, destination, date, weight, piece_count), create_booking(offer_id, shipment_details), track_shipment(awb_number), and more. A user authorizes the connection once, and Copilot stores the necessary credentials securely in an MCP configuration file. From that point forward, any prompt that implies cargo search or booking can trigger real API calls.

Because MCP runs locally or on a secure proxy, sensitive API keys never reach the AI model provider. Copilot’s runtime manages the tool execution, feeds results back as context, and always asks for confirmation before executing state-changing actions like booking. This design sidesteps the hallucination risks that plagued early AI plugins.

Microsoft added native MCP client support to Copilot in early 2026, starting with Copilot in Windows and later extending to Copilot for Microsoft 365. Windows users can configure MCP servers by editing a JSON file in %APPDATA%\\Copilot\\mcp_config.json or through a forthcoming settings UI. Several third-party tools like CargoMART are now available in the Copilot connector catalog, though enterprise admins can also deploy custom MCP servers across their organizations via Intune.

Two clicks from chat to air waybill

CargoAi’s integration is deceptively simple on the surface. A logistics operator might say:

“I need to ship 500 kg of automotive parts from Shanghai to Dallas. Show me options departing after 8 PM tonight.”

Copilot calls the CargoMART MCP server’s search tool, passing the parsed parameters. Within a few seconds, a table appears in the chat showing airlines, flight numbers, departure times, all-in rates, and available capacity. The operator can then click a button embedded in the response to proceed to booking.

Under the hood, the MCP server handles authentication, rate negotiation (CargoMART supports instant booking on many airlines), and returns structured data that Copilot renders as adaptive cards. Confirmations, AWB numbers, and tracking links arrive in the same thread. The entire workflow stays in the flow of conversation—no tabs, no copy-pasting, no manual rekeying of shipment data.

CargoAi supports over 3,000 trade lanes and integrates with 100+ airlines. Its CargoMART platform already processes tens of thousands of bookings monthly through traditional channels. Extending this to AI assistants opens the door for smaller forwarders who may not have the IT resources to build custom API integrations. They can now plug into the same capacity through a natural language interface.

Why Windows users should care

Copilot is evolving from a glorified chatbot into an agentic workbench. Microsoft’s Windows 11 2026 Update (version 26H2) identifies MCP connectors as a key pillar of the “Copilot as an orchestrator” vision. With the protocol in place, any Windows app that exposes an MCP server can become a Copilot skill, blurring the line between desktop software and AI assistance.

For logistics, which runs on Windows desktops in most freight forwarding offices, this means the tools they use daily—CargoWise, SAP TM, or CargoMART—can feed live data into Copilot without fragile screen scraping or brittle plug-ins. An MCP server is a lightweight service that can run on the same machine or on a company server, and Microsoft provides open-source SDKs in Python, .NET, and JavaScript to build one in an afternoon.

Security remains a priority. The MCP specification mandates explicit user consent for each tool invocation, and Copilot enforces that consent through its user interface. For enterprises, administrators can approve specific MCP servers and restrict which tools are available to which users, ensuring that only trained staff can create bookings while others might only search and track.

CargoAi’s server is open-source, available on GitHub under the MIT license. That transparency allows IT teams to audit the code, verify it only makes the API calls it claims, and even extend it with custom quoting logic or integration into internal ERP systems. Because the protocol is open, the same server works with ChatGPT Desktop and Claude Desktop—though Microsoft’s deep OS integration makes Copilot the most seamless experience for Windows shops.

The bigger picture: AI agents enter the supply chain

Freight tech has been notoriously slow to adopt consumer-grade UX, yet AI assistants are now pushing a new paradigm. Gartner predicts that by 2028, half of all transactional freight bookings will involve an AI intermediary. MCP connectors are the mechanism that allows domain-specific platforms like CargoMART to be that intermediary without reinventing the conversational wheel.

CargoAi is not alone. Several logistics marketplaces are building MCP servers: Freightos for ocean freight, TruckSmarter for FTL, and Flexport for end-to-end visibility. Microsoft itself has demonstrated MCP connectors for SAP, ServiceNow, and Salesforce. The pattern is clear: AI becomes the universal front end, and MCP is the standardized backplane.

For the air cargo industry still reeling from e-commerce volume surges and capacity crunches, faster quoting and booking translates directly into competitive advantage. A forwarder who can confirm a pharma shipment in 30 seconds instead of 30 minutes wins the business. Copilot plus CargoMART MCP makes that possible without any heavy IT lift.

There are limitations. MCP’s security model relies on local execution; running the server on a headless machine for team use requires careful proxying and authentication design. CargoAi’s current implementation expects one-to-one user-server pairing, though a multi-tenant version is in the works. Also, natural language is still fallible—ambiguous instructions like “cheapest” might ignore service-level requirements unless the model is prompted carefully. CargoAi includes safeguard prompts that ask clarifying questions before booking, but the industry will need to develop best practices.

Getting started with CargoMART on Copilot

For Windows users eager to try the integration:

  • Ensure you are running at least Windows 11 24H2 with the latest Copilot update (build 26100 or newer).
  • Open Copilot, go to Settings > Connectors, and search for “CargoMART” in the MCP catalog.
  • Click Install, and a configuration wizard will ask for your CargoMART API credentials. You can generate those from your CargoAi account under Developer Settings.
  • Once connected, you’ll see a list of available tools. Enable the ones you plan to use.
  • Start a new conversation and begin with a cargo search prompt, e.g., “Search CargoMART for flights from LAX to NRT tomorrow carrying 300kg of general cargo.”

Alternatively, power users can add the server manually by pasting its MCP endpoint into the mcp_config.json file. CargoAi’s GitHub repository provides a step-by-step guide and sample configuration.

For IT administrators, the connector can be deployed silently via Group Policy or Intune. The server itself is a small Node.js or Python process that can run on each user’s machine or on a central server that proxies requests. Microsoft’s MCP Gateway (in preview) offers an enterprise-grade option that adds load balancing, auditing, and role-based access control.

What’s next

CargoAi plans to expand the MCP server’s capabilities to include contract rate negotiation, automatic rebooking when flights are delayed, and carbon-offset calculation. On Microsoft’s side, future Copilot updates will allow chaining multiple MCP tools into a single agentic workflow—imagine asking Copilot to “find cargo, book a truck to the airport, and update our TMS,” all in one go.

The air cargo desk of the future may have no desk at all—just a Windows tablet running Copilot, a CargoMART MCP connector, and a human expert guiding the AI with domain knowledge. CargoAi’s move on June 5, 2026, marks a tangible step toward that vision.