We’re on the cusp of a new AI-powered world, where agents automatically execute a range of tasks previously handled by humans. But how will we connect all of these agents with data and coordinate with AI tools? That’s the role of a new protocol unveiled by Anthropic several months ago called Model Context Protocol, or MCP.
Anthropic launched MCP last fall with the goal of providing an open standard for connecting large language model (LLM)-powered AI agents with the data and the tools they need to perform their tasks. “Instead of maintaining separate connectors for each data source, developers can now build against a standard protocol,” the AI company said in a November 25 blog post announcing MCP.
Anthropic designed MCP as a lightweight architecture to enable developers to build secure, two-way connections between the MCP server (data sources) and LLM-powered apps utilizing the MCP client. The goal is to go beyond basic AI workflows, such as inference and retrieval-augmented generation (RAG), to enable to be taken beyond the AI application itself, such as automatically sending an email through Salesforce or booking a rental through Airbnb.
“Think of MCP like a USB-C port for AI applications,” Anthropic wrote in an introduction to MCP. “Just as USB-C provides a standardized way to connect your devices to various peripherals and accessories, MCP provides a standardized way to connect AI models to different data sources and tools.”
Image courtesy Mahesh Murag, Anthropic
Anthropic has published specifications and shared documentation for MCP, which uses JSON-RPC 2.0 messages to establish stateful connections between LLM hosts (and the MCP client) and the remote data source (the MCP server). It has also developed software development kits (SDKs) to enable developers to build MCP clients in Python, Java, C#, TypeScript, and Kotlin. It’s also shared pre-built MCP servers for popular enterprise systems like Google Drive, Slack, GitHub, Git, Postgres, and Puppeteer.
When Anthropic launched MCP in November, it already had the backing of several major players, including AI development tool makers Codeium, Replit, Sourcegraph, and Zed. It also touted how Block (the e-payment company formerly known as Square) has adopted MCP to “remove the burden of the mechanical so people can focus on the creative,” as Block CTO Dhanji Prasanna put it.
As momentum builds behind agentic AI, MCP appears to be gaining a foothold in the broader AI ecosystem.
“It does feel like it is leading the way,” said Anupam Datta, principal research scientist at Snowflake and the lead for the Snowflake AI Research Team. “I think there’s a lot of promise there.”
As agentic AI takes off, companies will need a way to ensure interoperability of the AI agents, AI tools, and data sources across different platforms, Datta told BigDATAwire at the recent GTC 2025 conference in San Jose, California.
“I think the problem that it’s working on, that MCP is trying to address, is a very real problem,” Datta said. “We are starting to explore MCP. The field is still early, so it’s hard to say if this becomes the standard.”
MinIO, which develops an S3-compatible object store, has enthusiastically boarded the MCP train. Last week, the company announced support for MCP in AIStor, the enterprise version of its open source object storage system. Specifically, MinIO announced a preview for an MCP server that will open AIStor up to AI agents. MinIO’s MCP server supports 25 different commands in AIstore. It also announced that it’s building administration capabilities into its MCP offering to prevent unauthorized access from MCP clients, as well as introducing monitoring capabilities for MCP server.
“Now you can talk to MinIO data through the MCP server,” MinIO co-founder and co-CEO AB Periasamy said. “It will discover what you are looking for. It basically kind of advertises the data it has. And then it gives you access to the data.”
For instance, say you wanted to send an Uber to pick up your child. You could tell an AI agent, such as Amazon’s Alexa+, to do that, but it would likely run into problems around who exactly the person is, and what exact Uber account you’re referencing. MCP will resolve those problems, Periasamy said.
“It’s not like AI doesn’t know how to do this,” he told BigDATAwire at the recent GTC 25 conference. “It absolutely knows how to do this. But AI doesn’t have access to your Uber account. It doesn’t know who your kid is. It knows how to get this done, but it doesn’t have any context. That’s model context protocol.”
Another early MCP adopter is the AI firm Yurts, which develops tools to help customers integrate AI into their enterprise applications. “It’s hard to say [if it will become the standard],” said Ben Van Roo, the company’s co-founder and CEO. “But I think they’ve taken a thoughtful approach.
Nvidia CEO Jensen Huang says agentic AI will drive a 100x increase in inference workloads
The 50-person company is developing software to help large organizations deploy generative AI, LLMs, and AI agents into mature enterprise environments. Its software helps to monitor and manage AI apps, including copilots and AI agents. Yurts had started developing its own protocol before Anthropic launched MCP.
“When you’re a really small company, you have to choose what bets you’re not going to make,” Van Roo told BigDATAwire recently. “And as soon as it came out, it was like, we’ve already built a bunch of the infrastructure, but let’s take advantage of their process. They’ve done a really nice job.”
Other vendors are also building upon MCP. For instance, Alation announced support for MCP as part of its Agentic Platform earlier this month. The company has adopted MCP within its AI Agent SDK to help ensure that any data accessed through an AI agent workflow is abiding by the company’s established data governance policies.
Cloudflare also announced a new remote MCP service that enables less technical users to build and deploy MCP servers for services such as oAuth authentication. “Remote MCP support is like the transition from desktop software to web-based software,” Cloudflare engineers wrote in a March 25 blog post. “People expect to continue tasks across devices and to login and have things just work. Local MCP is great for developers, but remote MCP connections are the missing piece to reach everyone on the Internet.”
The NoSQL database company DataStax, which is in the process of being acquired by IBM, is adopting MCP with its hosted database service. With its new Astra DB over MCP service, customers can enable AI agents, such as Anthopric’s Claude Desktop, to tap into data stored in Astra DB.
“From here, you can ask Claude to do anything you like inside your database: create collections, insert data, clean up, and more,” Tejas Kumar, DataStax’s developer relations engineer, wrote in a blog post. “This is a handy way of interacting with your database via an AI assistant but we can do more when we use Cursor as our MCP client.” (Cursor, by the way, is “an AI-enabled version of VS Code that integrates MCP tools directly into your development workflow.)
MCP clearly has the early lead in the race to develop a communication standard for agentic AI. Unless another AI giant proposes a competing protocol soon, MCP may win by simply being first.
Related Items:
Can You Afford to Run Agentic AI in the Cloud?
The Future of AI Agents is Event-Driven
What Benchmarks Say About Agentic AI’s Coding Potential
The post Will Model Context Protocol (MCP) Become the Standard for Agentic AI? appeared first on BigDATAwire.
Leave a Reply