MCP is Not That New

Why is everyone talking about MCP? It's powered by AI tooling and function calling. Is it just a simple abstraction of server/AI logic? If so what gives? MCP servers (Model Context Protocol servers) share many similarities with RESTful APIs in both technical implementation and conceptual use. As Anthropic (the creator of MCP) describes, MCP is a “universal, open standard for connecting AI systems with data sources, replacing fragmented integrations with a single protocol”. This closely parallels the role of good ol' REST APIs in web development, where disparate services are unified under common "agreements". Maybe following an old proven concept is not a bad thing. Nevertheless, the power of the protocol is something we've seen before like stateless interactions, standardized I/O formats, clear endpoint logic, and modular integration patterns. REST AND MCP Both REST APIs and MCP servers are designed for stateless, request-response style interactions. In REST, each HTTP request is independent – the server does not retain client state between calls. It's an input output situation. Similarly, MCP calls are handled in isolation by default, with any needed context passed explicitly in the request. That is not saying you can't have something within the MCP that uses context in it's own request but ideally right now you would not have context across requests. Right now an MCP client can query a log resource by providing a URI and parameters and get back the result without the server “remembering” prior queries. This means that if an AI agent needs information or an action, it formulates a self-contained request, and the MCP server returns a response – no session or history required (unless explicitly implemented). This stateless design simplifies composition of multiple calls and aligns with the stateless nature of RESTful services. Simply Why is it Useful? I could write paragraphs but serious just 2 points: Standardization REST typically uses JSON over HTTP, and MCP likewise uses a structured JSON-based protocol – in fact, MCP leverages JSON-RPC 2.0 messages under the hood for communication. AI Agents Another plus, an AI agent can automatically detect available MCP servers and their functions via the protocol’s standardized interface Sign Me Up! Where can I start? Developers and companies are increasingly relying on such marketplaces to discover and integrate new MCP servers into their AI workflows, paralleling how developers use platforms like RapidAPI or AWS Marketplace to find ready-made REST APIs. Below are some real-world examples of MCP server marketplaces and how they’re being used: Cline’s MCP Marketplace – Cline (an AI agent/assistant platform) offers a curated marketplace of MCP servers built right into its interface. When a developer chooses a server from Cline’s marketplace, the tool will automatically clone and set it up, instantly extending the AI’s capabilities. I gotta try this to be honest. Mintlify’s MCP Registry – I came across this one but yet to try it. Maybe I will dive into it but did not look as good as Cline or the next one... SMITHERY.AI The existence of thousands of MCP endpoints on Smithery underscores the explosive community growth – many contributors are creating servers for all sorts of services (from database management to marketing tools), and a marketplace like this organizes them. This is really sick, can't wait to try it. What's next Rob? I am going to post a project or integrate a MCP that seems useful next week, stay tuned! As always subscribe.

Mar 25, 2025 - 17:01

Why is everyone talking about MCP? It's powered by AI tooling and function calling. Is it just a simple abstraction of server/AI logic? If so what gives?

MCP servers (Model Context Protocol servers) share many similarities with RESTful APIs in both technical implementation and conceptual use. As Anthropic (the creator of MCP) describes, MCP is a “universal, open standard for connecting AI systems with data sources, replacing fragmented integrations with a single protocol”. This closely parallels the role of good ol' REST APIs in web development, where disparate services are unified under common "agreements". Maybe following an old proven concept is not a bad thing. Nevertheless, the power of the protocol is something we've seen before like stateless interactions, standardized I/O formats, clear endpoint logic, and modular integration patterns.

REST AND MCP

Both REST APIs and MCP servers are designed for stateless, request-response style interactions. In REST, each HTTP request is independent – the server does not retain client state between calls. It's an input output situation. Similarly, MCP calls are handled in isolation by default, with any needed context passed explicitly in the request. That is not saying you can't have something within the MCP that uses context in it's own request but ideally right now you would not have context across requests. Right now an MCP client can query a log resource by providing a URI and parameters and get back the result without the server “remembering” prior queries. This means that if an AI agent needs information or an action, it formulates a self-contained request, and the MCP server returns a response – no session or history required (unless explicitly implemented). This stateless design simplifies composition of multiple calls and aligns with the stateless nature of RESTful services.

Simply Why is it Useful?

I could write paragraphs but serious just 2 points:

Standardization REST typically uses JSON over HTTP, and MCP likewise uses a structured JSON-based protocol – in fact, MCP leverages JSON-RPC 2.0 messages under the hood for communication.
AI Agents Another plus, an AI agent can automatically detect available MCP servers and their functions via the protocol’s standardized interface

Sign Me Up! Where can I start?

Developers and companies are increasingly relying on such marketplaces to discover and integrate new MCP servers into their AI workflows, paralleling how developers use platforms like RapidAPI or AWS Marketplace to find ready-made REST APIs. Below are some real-world examples of MCP server marketplaces and how they’re being used:

Cline’s MCP Marketplace – Cline (an AI agent/assistant platform) offers a curated marketplace of MCP servers built right into its interface. When a developer chooses a server from Cline’s marketplace, the tool will automatically clone and set it up, instantly extending the AI’s capabilities. I gotta try this to be honest.

Mintlify’s MCP Registry – I came across this one but yet to try it. Maybe I will dive into it but did not look as good as Cline or the next one...

SMITHERY.AI
The existence of thousands of MCP endpoints on Smithery underscores the explosive community growth – many contributors are creating servers for all sorts of services (from database management to marketing tools), and a marketplace like this organizes them. This is really sick, can't wait to try it.

What's next Rob?
I am going to post a project or integrate a MCP that seems useful next week, stay tuned!

As always subscribe.