There may be turn-key setups available that would allow you to focus on some specific aspect, like MCP. (If there aren't that could be an interesting project! 😉)
My impression is that coding an agent and one or more MCPs to supply it is a substantial endeavor. I'd look for an angle before I began such a project, like track down a school program/bootcamp/tutorial that presents a similar project framed like "Develop this AI example as a job search calling card."
I might have created as many as a thousand or more interfaces, though. I tended to gravitate to software interoperability, connecting one program or system to others. Sometimes this was a client-server arrangement like using a database, while at other times the service producer/consumer model was more relevant.
The only concept in MCP that would be novel to my background is the (AI) autonomous agent. Everything I read about the MCP itself sounds well-established and familiar, if the introspection can be on the obscure side. By obscure I want to convey something about use-case: usually a system has a very limited scope and excessive abstraction can be a hindrance for maintenance. Like, my HVAC API design can presume users intending to access environmental controls. Adding introspection, the ability to advertise a service that can be discovered by a naive client, is probably just adding points of failure for no value gained...
Anyway, have you developed a MCP? Share how it went and what you discovered!
A project I was thinking about was around healthcare. More like effeciencies in medical billing.
I have not yet tried to create an AI agent using MCP. Was curious if it's even worth the trouble to learn how.
I have not 😁
I might have created as many as a thousand or more interfaces, though. I tended to gravitate to software interoperability, connecting one program or system to others. Sometimes this was a client-server arrangement like using a database, while at other times the service producer/consumer model was more relevant.
The only concept in MCP that would be novel to my background is the (AI) autonomous agent. Everything I read about the MCP itself sounds well-established and familiar, if the introspection can be on the obscure side. By obscure I want to convey something about use-case: usually a system has a very limited scope and excessive abstraction can be a hindrance for maintenance. Like, my HVAC API design can presume users intending to access environmental controls. Adding introspection, the ability to advertise a service that can be discovered by a naive client, is probably just adding points of failure for no value gained...
Anyway, have you developed a MCP? Share how it went and what you discovered!