Model Context Protocol (MCP) has quickly emerged as a topic of interest in conversations about AI for business. This post explains the significance of MCP, how the technology works, its revolutionary aspects, and how organizations can position themselves to use it.
MCP was released by Anthropic last November, described as “a new standard for connecting AI assistants to the systems where data lives, including content repositories, business tools, and development environments.”1
MCP is significant because it directly addresses a problem that AI agents present for most organizations: interconnectivity. In an ideal framework, AI agents can be sequenced to work together while utilizing shared tools and information. This is often described as orchestration, and in that sense, an MCP is like the sheet music that gets passed around to all of the players in the orchestra before a concert.
Sheet music comes in a standardized language that the musicians in an orchestra can understand. Each player’s set of instructions is different, telling them which instrument to use, when to use it, and how. In this way, MCP can make it far easier for AI agents to communicate with the other elements in an organization’s technology ecosystem.
How does MCP work?
As a protocol with standardized ways to communicate information, MCP gives AI agents clear rules for how to locate, connect to, and use external tools. In action, an AI agent uses JSON (JavaScript Object Notation) to query an MCP server that provides access to requested tools, resources, and prompts. This provides two-way communication between AI agents, data sources, and tools.
MCP servers exist in an open-source repository, and Anthropic has shared pre-built servers for enterprise systems like Google Drive, Slack, GitHub, Git, Postgres, and Puppeteer. Because MCP is open-source, it’s technology agnostic, and anyone can experiment with it using their own tools and models.
Advantages of MCP include:
- More flexible and scalable than custom API integrations,
- Compatible with frameworks like LangChain and Agents,
- Compatible with an open technology ecosystem that integrates market-best tools and models.
“Without MCP (or something like it), every time an agent needs to do something in the world — whether fetching a file, querying a database, or invoking an API — developers would have to wire up a custom integration or use ad-hoc solutions,” Ksenia Se wrote in her post for Hugging Face. “That’s like building a robot but having to custom-craft each finger to grasp different objects — tedious and not scalable.”2
Does MCP change everything?
Providing a standardized interface for AI agents to communicate with a broader ecosystem of information and software is revolutionary. Still, the initial release of MCP went largely unnoticed until earlier this year, when MCP seemed to eclipse AI agents as the focal point of marketplace attention.
“MCP is bigger as an idea than it is as an actual technological achievement,” says Robb Wilson, CEO and co-founder of OneReach.ai, noting that the real revolution comes with the trajectory MCP opens. “Its implications and where it’s going is what’s exciting.”
Those implications relate to how the MCP closes the gap between LLM-based AI agents and real-world business systems and information. Block (Square), Apollo, Zed, Replit, Codeium, and Sourcegraph were early MCP adopters, and the ecosystem now has more than 1,000 community-built MCP servers.
Add to this growth in the MCP ecosystem Sam Altman’s announcement last month that OpenAI will support MCP across its products, including the desktop app for ChatGPT.3 Just days ago, Google released their own Agent2Agent (A2A) protocol that they describe as a complement to MCP. The tech giant cited support from 50+ partners, including Atlassian, Intuit, PayPal, Salesforce, ServiceNow, Workday, and leading service providers like Accenture, BCG, Capgemini, Cognizant, Deloitte, McKinsey, and PwC.4
This surge of interest and activity is noteworthy, but it points to something bigger than a single protocol. With traditional software, various tools and features are bundled by graphical user interfaces (GUIs). Agentic AI puts us on the cusp of a world where anyone can turn to a piece of technology and simply ask for help. Behind the scenes, a flurry of activity that includes MCP (or something like it) brings back the information or action requested. In this scenario, the GUI software bundle loses all relevance.
As Wilson suggests, MCP breaks software into pieces: tools for hire that users will only care about in the moment that they are needed.
“What we’re talking about is a single UI for all our software. That’s massive. If we’re talking about one UI that you can use to get a bunch of stuff done, people are going to want to own that UI. OpenAI thinks and hopes they will, Anthropic hopes and thinks they will.”
How to leverage MCP
Wondering who might end up owning a lone UI perched high on a distant mountaintop clearly isn’t top-of-mind for businesses at this moment, but it does point to a critical factor that organizations have to consider as they are assembling a framework for agentic AI. The alternative to one UI for all software is individual organizations with UIs that are connected to their unique software ecosystem.
The backend of a truly dynamic and useful organizational UI needs to be both open to new technologies and flexible enough to rearrange itself around any new requirements that come with them. MCP is open and model-agnostic, which aligns with these requirements, but its sudden rise to prominence is also a reminder that in this new era of conversational technologies, the idea of “market-best” is completely fluid. Revolutionary tools and approaches will continue to erupt and stumble over one another as agent orchestration matures.
A world filled with high-functioning tech ecosystems might seem like a distant promise, but the race toward them is already underway. MCP is a key piece in this journey, both in the way that it standardizes communication between machines and in the way that it can contribute to an open ecosystem, where any tool or data source can become part of a bigger process automation.
- 1“Introducing the Model Context Protocol,” Anthropic
- 2Ksenia Se, “#14: What Is MCP, and Why Is Everyone – Suddenly! – Talking About It?,” Hugging Face
- 3Kyle Wiggers, “OpenAI adopts rival Anthropic’s standard for connecting AI models to data,” TechCrunch
- 4“Announcing the Agent2Agent Protocol (A2A),” Google
The article originally appeared on OneReach.ai.
Featured image courtesy: Pawel Czerwinski.
