In a recent LinkedIn post, Luís Rodrigues delves into the often-confused relationship between Message Queue Protocol (MCP) and Application Programming Interfaces (APIs), arguing that they serve distinct yet complementary roles in modern AI agent systems. Rodrigues emphasizes that comparing MCP directly to an API is a fundamental misunderstanding of their functions.
Deconstructing the API
Luís Rodrigues begins by outlining the typical workflow of an API, highlighting the manual effort involved in its creation and maintenance. He explains the sequential nature of API interactions:
“Your app fires an HTTP request. The gateway checks the token, decides where it goes. A service picks it up, runs the query, and hands back JSON. The screen updates. Someone wrote every step of that path by hand, and someone maintains it.”
According to Rodrigues, this process assumes a pre-existing knowledge on the part of the caller regarding what specific queries can be made. It represents a defined contract between two systems that have prior knowledge of each other.
Understanding MCP’s Role
In contrast, Rodrigues describes the MCP process as a more dynamic discovery mechanism. He illustrates this with an agent-based workflow:
“An agent gets a task. It looks at what servers are connected and reads what each one says it can do. It picks what makes sense and sends a request in a shape it already knows. The answer comes back the same way. Nobody wrote a custom path. The agent found it.”
As Luís Rodrigues points out, MCP’s strength lies in its ability to allow a system to discover available functionalities without prior explicit programming for each interaction. He frames this as:
“One is a contract between two systems that were introduced beforehand. The other is a way for a system to walk into a room and ask what’s available.”
The Complementary Nature of MCP and API
Rodrigues strongly refutes the idea that MCP replaces APIs. He argues that the core functionality often still relies on APIs, even when an MCP layer is present. He uses the analogy of the MCP server sitting in front of common tools like Postgres, GitHub, and Drive, interacting with them through their existing endpoints – the same endpoints an application would use.
MCP as a Translation Layer
Luís Rodrigues defines MCP as a translation layer specifically for AI models. “MCP is a translation layer for models. Nothing more, nothing less,” he states. This layer facilitates the interaction between AI agents and various services by abstracting the complexities of direct API calls and enabling dynamic discovery of capabilities.
He concludes by relating this back to practical development, suggesting that developers who have created numerous wrappers for tools used by agents will understand the underlying necessity and benefit of MCP. Rodrigues then prompts his audience by asking, “What was the first tool you connected over MCP?”
Through this explanation, Luís Rodrigues clarifies that MCP and API are not competing technologies but rather different layers in a system, with MCP enabling more flexible and discoverable interactions for AI agents, while APIs continue to perform the fundamental work of data exchange.
📝 About This Content
This article is based on insights shared by Luís Rodrigues on LinkedIn.
📅 Originally posted on August 15, 2026 | View original post on LinkedIn →