Some big ideas to round out the week.
The New Stack with the MCP Server’s first year review. In its first year, the Model Context Protocol (MCP) from Anthropic has seen significant adoption, serving over 278 million requests across more than 5,000 organizations. This protocol aims to provide a universal standard for integrating AI systems with data sources, thereby addressing fragmentation in the tech landscape. Notably, companies like Cloudflare and Vercel have developed tools to facilitate MCP server creation, enhancing functionality and user experience. Despite initial challenges, including a rough start with various integration issues, the MCP has evolved to become a critical component of any large language model client. Observations from developers indicate a shift away from outdated technologies, with a growing preference for HTTP Streamable over Server-Sent Events, reflecting an industry trend towards more robust solutions. As the MCP continues to mature, experts anticipate further enhancements in tool consolidation and OAuth support, underscoring its potential as a key player in AI integration moving forward.
And Information Week with a look at deploying them – building is easy, using.. less so. According to Anand Chandrasekaran, principal engineer at Arya Health, while the initial setup is easy, maintaining operational success is complex and fraught with risks, as swift implementation can lead to exploitation vulnerabilities. Experts highlight that MCP’s promise for enterprises is tempered by its current lack of readiness for large-scale deployment. Mohith Shrivastava, principal developer advocate at Salesforce, notes the hurdles in transitioning from proof-of-concept to live systems, emphasizing that security, governance, and orchestration are critical for operational success. The article outlines five main challenges in implementing MCP, such as security risks, tool overload, scaling issues, production gaps, and compliance concerns, stressing the importance of establishing robust governance frameworks to mitigate these risks effectively.
Why do we care?
MCP is taking off because it makes connecting AI to systems dead simple. And that’s the problem. Building is easy — governing it is not. We’ve seen this movie before with APIs, with SaaS sprawl, with RPA. Fast adoption creates a mess if nobody’s thinking about boundaries.
Right now, MCP is developer-friendly but not enterprise-ready. The security model’s thin, the governance story is vague, and companies are spinning up these integrations way faster than they’re securing them. That’s how shadow automation creeps in — not because anyone’s being malicious, but because the tools make it too easy.
For MSPs, this is where you step in. Don’t get pulled into building random MCP servers. Focus on the discipline: Who owns this integration? What data is flowing? Where’s the authentication coming from? Who shuts it down when it breaks? Those questions matter more than the tech.
The winners in this next wave aren’t the ones writing little automations — it’s the ones who build the guardrails so customers don’t blow themselves up. MCP is going to matter, but only if someone is managing the chaos. That someone should be you.