The world has revolved around B2B (business to business) sales for years. It’s been the model ever since I’ve been in the industry. However many are predicting an upcoming shift from B2B to B2A. B2A means business to agents.
Eventually agents will connect to tools or businesses and be able to perform tasks. So instead of a business selling software to help a human to complete a task the business will connect to a human’s AI agent. For example, imagine a business selling software for vulnerability management. In B2B world a security engineer buys software that helps them find and remediate vulnerabilities. In a B2A world an agent would access CVE data on vulnerabilities from a business and cross reference that with software dependancies found in a github repo and then remediate vulnerabilities on its own.
The B2A world relies on agents being able to securely connect to businesses and resources. The Model Context Protocol (MCP) developed by Anthropic is a proposed standard that’s gaining traction that is well positioned to be the protocol that enables B2A. This protocol is used to create MCP servers that federate access to data on databases, applications, or tools. MCP servers have 12.7k stars on Github and 1000+ open source community MCP servers have been built.
Others have started building support for MCP in response to the growing popularity. On February 27th Langchain released MCP adapters for LangGraph that provides a way for agents built on LangGraph to easily connect to MCP servers. You can see an example on github.
Last month at the AI Engineer Summit in NYC, Mahesh Murag, Applied AI at Anthropic at AI, presented a workshop on MCP. In his words, MCP standardizes how AI Applications interact with external systems such as prompts, tools, and resources.
He shared how Anthropic saw the same problem over and over again. Each AI application would need to write separate implementations for how their AI applications talk to external resources. Murag called this an n*n problem. I’ve illustrated that in the image below.
Instead of n AI Apps each creating the same n integration for the same service, the service can create one MCP server that each AI app can easily connect to. So it’s many to one but in the opposite direction.
MCP will be the foundational protocol for agents. Greg Isenberg tweeted on March 1st about how agents will disrupt current software. He said “the winners will be companies that expose their software's capabilities through agent-friendly APIs and position themselves as the most trustworthy information sources and execution engines in their domain.” MCP has a foundational role to play in helping to connect trustworthy information to AI applications. Greg gives the example that with agents “someone can say "analyze our Q2 performance" rather than clicking through Tableau, or "optimize our ad campaigns" instead of navigating Meta's ad manager.” This example would be possible with MCP servers for Tableau or Meta’s ad manager.
Here’s a couple of cool MCP servers in action. See one that Vanta built to allow LLMs to connect to their auditor API. With this you could use an LLM connected to this MCP server to use natural language to ask what recent audits have been done for your organization.
See another that Graphlit, a knowledge API service, built that allows AI apps to connect to Graphlit.
Business to agents will likely be the future and MCP will be a big part of that. I’m excited to see what you will build with MCP servers. Growth Cyber is here for you to help you think through AI app design or agentic threat modeling.