MCP Adoption Surges as Enterprises Weigh New Security Risks

A Protocol Goes Mainstream

Model Context Protocol (MCP), the open standard that lets large language models connect directly to external tools, data sources, and business systems, has moved from a niche developer curiosity to a foundational piece of enterprise AI infrastructure in a remarkably short time. Originally introduced by Anthropic, MCP essentially gives AI agents a standardized way to "call" software services — translating user intent into structured backend actions rather than forcing developers to write bespoke integrations for every model and every API 1. That framing helps explain why the protocol is increasingly described as one of the most important pieces of plumbing in the current AI agent boom 6.

Who's Building on It

The rush to adopt MCP spans media, finance, advertising, and social platforms. Reuters launched its own MCP server so that news organizations and enterprise customers can pull trusted, licensed news content directly into their agentic AI workflows, positioning trustworthy data as a differentiator in an environment full of AI-generated noise 7. Snap rolled out an MCP server for advertisers, letting marketers connect third-party AI tools to its ad platform for what the company calls more "human-first" campaign guidance 3. X similarly launched a hosted MCP server to make it simpler for developers to plug AI applications into its API 8. In banking, Grasshopper Bank entered a private beta using Narmi's MCP server, aiming to let business clients query financial insights through Anthropic's Claude assistant — an early sign of MCP moving into regulated, customer-facing financial workflows 4.

Guidance for Builders, and Growing Pains for the Protocol

As adoption spreads, product leaders are being urged to think strategically about where MCP fits into their roadmaps, with guidance emphasizing careful evaluation of use cases, data governance, and integration priorities before wiring agents into core systems 2. The protocol itself is also evolving: a recent specification overhaul makes session handling "stateless" on the server side, mirroring how conventional websites already operate, in an effort to simplify implementation for developers 6.

Security Emerges as the Central Concern

That same overhaul, however, has shifted significant security responsibility away from the protocol and onto individual developers and platform operators, raising fresh concerns about consistency and oversight across implementations 9. The stakes are underscored by a reported surge of more than 40 CVEs tied to MCP in just four months, a pace that analysts say has turned MCP security from a technical footnote into a board-level risk issue as LLMs move from experimental sandboxes into mission-critical enterprise deployments 5.

Why It Matters

Taken together, the coverage points to a protocol maturing faster than the guardrails around it. MCP is becoming the default way enterprises connect autonomous AI agents to real business systems — news feeds, ad platforms, social APIs, and banking data alike — but its rapid, decentralized adoption means security practices are still catching up to the scale of deployment.