How an API integration works
Say you want your agent to plan trips. With APIs, you would sign up for a flight provider, a hotel provider, and a car rental provider. Each one has its own authentication, request format, and response shape. You write code to call each one, parse the results, handle errors, and turn everything into tool definitions your agent can use. When a provider changes its API, you update your code.
That work is fine for a traditional app. For an agent, it is a lot of plumbing before you get to the part users care about.
How MCP works
With MCP, your agent connects to an MCP server through one endpoint. The server publishes a list of tools, each with a plain description and a defined set of inputs. The agent reads that list, decides which tool fits the request, sends the inputs, and gets back structured results it can reason over.
MCP was introduced by Anthropic in late 2024 and is now supported by Claude, ChatGPT, Cursor, and most major agent frameworks.
// Request: the agent asks the server what it offers { "jsonrpc": "2.0", "id": 1, "method": "tools/list" } // Response: the shape of each tool entry { "jsonrpc": "2.0", "id": 1, "result": { "tools": [ { "name": "<tool_name>", "description": "<what the tool does, in plain language>", "inputSchema": { "type": "object", "properties": { … } } } ] } }
MCP vs API at a glance
| Feature | Travel API | MCP server |
|---|---|---|
| Built for | Developers writing code | AI agents using tools |
| Setup | One integration per provider | One connection for every tool on the server |
| How the agent learns what’s available | You write the tool definitions | The server describes its own tools |
| Adding a new travel vertical | New integration | Already there if the server offers it |
| Response format | Different per provider | One consistent schema |
Is MCP replacing APIs?
No. Most MCP servers run on top of APIs. The difference is who does the integration work. RouteStack’s MCP endpoint is backed by direct supplier connections that Custom Travel Solutions has built over more than 30 years. You don’t integrate those suppliers yourself. Your agent connects once and gets the tools.
What this looks like with RouteStack
A traveler asks your agent: “Find me a nonstop from San Francisco to New York on March 12 and a hotel near Midtown under $300 a night.”
Your agent calls RouteStack’s flight search tool and hotel search tool through the same connection. Both return live pricing and availability in the same JSON format, so the agent can compare and present options right away. When the traveler books, you earn commission, paid weekly via ACH.
RouteStack covers flights, hotels, cars, activities, and vacation rentals through one hosted MCP endpoint.
FAQ
Is MCP just another API?
MCP runs over standard web technology, but it is a protocol designed for AI agents. Every MCP server describes its tools in the same way, so any MCP-compatible agent can use them without custom code.
Do I still need credentials?
Yes. You connect to RouteStack with a token. The sandbox token is free, and you swap it for a production token when you go live.
Which AI clients support MCP?
Claude, ChatGPT, Cursor, and most agent frameworks support MCP servers, including remote servers like RouteStack.
Can I test before paying?
Yes. Every account includes an unlimited sandbox with real travel inventory from popular destinations, plus 1,000 free tokens a month on the Hiker plan.
Start building free.
Connect your agent to RouteStack’s sandbox in about a minute.