How MCP (or similar standards) Could Change the Way We Use Software with AI#
I think many of my colleagues here at Dinero will agree when I say Iโve been deep down the MCP-server rabbit hole lately ๐ณ๏ธ๐
Iโve spent quite a bit of time trying to understand MCP servers, public opinions around the standard, andโmost importantlyโhow it might shape the way we use software in the future.
And now many of you might be asking:
What the heck is an MCP server? ๐ค
Hereโs my best explanation so far:
MCP is a standardized frontend for AI agents.
Iโve created a diagram to illustrate how I imagine MCP servers might impact the future of software usage.
Below is a brief walkthrough of the diagram ๐
๐ง๐ผ๐ฑ๐ฎ๐#
Users interact manually with each system (Salesforce, HubSpot, Zendesk) via individual frontends. Everything relies on human contextual understanding and manual navigation.
๐ฃ๐ผ๐๐ฒ๐ป๐๐ถ๐ฎ๐น ๐๐๐๐๐ฟ๐ฒ ๐ญ#
Users talk to a single AI agent, which uses MCP servers to interact with various platforms. No more jumping between systems. The AI understands the task, and the MCP servers ensure the AI agent can interface with and understand each system.
๐ฃ๐ผ๐๐ฒ๐ป๐๐ถ๐ฎ๐น ๐๐๐๐๐ฟ๐ฒ ๐ฎ#
Users still only interact with one AI agent. However, this agent communicates with an MCP core, which delegates tasks to specialized AI agents for each system. This enables an even more intelligent and modular approach, where tasks are handled by experts (AI agents) for each systemโlike a team of specialists orchestrated behind the scenes.
If you buy into this, several interesting questions remain in the โPotential Future 2โ scenario:
Who will build and own the AI agent broker?
Will it be controlled by a single company or developed collaboratively by multiple players?
Will it be open-source or proprietary?
Will software companies in the future need to build an AI agent interface for their productsโjust as they do today for desktop and mobile?
Whatโs the incentive for companies to contribute agents to a shared broker?
Who decides which agent gets delegated a taskโand on what basis?
Will brokers develop โpreferencesโ or bias towards certain agents?