When Anthropic unveiled the Model Context Protocol (MCP) in mid-2024, it was quickly compared to REST for AI. MCP was pitched as a connective fabric that could make agents and large language models (LLMs) more useful by standardizing how they access external tools, data, and services.
For developers, MCP promises interoperability. For enterprises, it offers a vision of composability: AI systems that don’t live in silos but can plug into workflows, apps, and knowledge repositories.
But the narrative so far has been one-dimensional. Most of the conversation has focused on functionality (what MCP enables). Far less attention has been paid to security and governance (what MCP endangers if left unchecked).
That oversight matters because without robust guardrails, MCP could follow a trajectory similar to early cloud adoption: rapid growth shadowed by uncontrolled risk exposure.
Most importantly, as described by Randy Bias, MCP will not cross the enterprise adoption chasm unless it matures into a secure-by-design protocol.
MCP in Context: What It Solves
LLMs are powerful but inherently limited by the scope of their training data and context window. To make them practical, they need the ability to call out to fetch data, trigger actions, and interact with business logic.
MCP defines a standard way to describe these connections and make them consumable by models or agent frameworks.
Think of it as the API layer for AI agents. Just as REST and GraphQL became the lingua franca of web services, MCP is being positioned as the universal interface for model-to-environment interactions.
- Developers see MCP as a productivity boost: build once, connect everywhere.
- Vendors see it as a path to interoperability: an ecosystem that avoids lock-in.
- Enterprises see the potential to create agentic workflows that span tools, clouds, and departments.
It’s no surprise that momentum is building. But adoption curves aren’t just about utility. History shows that standards only achieve mass enterprise uptake once they demonstrate both capability and trust.
The Overlooked Challenge: Security and Governance
MCP doesn’t just connect agents to APIs; it potentially grants autonomous systems direct access to sensitive corporate functions. That makes it far more risky than REST ever was.
Consider a few scenarios:
- A finance agent using MCP could execute trades or access payroll systems.
- A healthcare agent could query EHR systems, exposing HIPAA-regulated data.
- A customer service agent could trigger refunds, escalations, or account changes at scale.
Without rigorous access controls, audit trails, and policy enforcement, enterprises risk creating what Mirantis calls a “shadow agent” problem. This echoes the shadow IT explosion of the early cloud era. Developers, eager to innovate, could spin up MCP-enabled agents that bypass governance frameworks entirely.
The result? Invisible risk at machine speed.
Lessons from the Past: REST, Cloud, and Shadow IT
The analogy to REST is helpful but incomplete. REST succeeded not just because it was simple, but because it was securable as it was bolted onto existing identity, encryption, and API gateway infrastructure. It emerged alongside OAuth, JWT, and the API management ecosystem that enterprises could wrap governance around.
Cloud adoption followed a similar arc. In the mid-2000s, AWS made it simple to spin up compute instances. Enterprises raced to the cloud, but soon faced shadow IT headaches: employees provisioning infrastructure without oversight, creating compliance and cost problems. The solution was governance tooling: cloud management platforms, CSPM, FinOps.
MCP is at the equivalent of AWS 2007. Powerful. Exciting. But if it grows faster than its security stack, the risks will metastasize before the benefits are fully realized.
What’s at Stake: The Enterprise Adoption Gap
For MCP to make the leap from developer darling to enterprise staple, it has to prove it can operate safely in regulated, high-stakes environments. That means answering a few critical questions:
- Authentication & Authorization – How do we ensure only trusted agents can invoke sensitive MCP connections?
- Policy Enforcement – How do we prevent agents from escalating privileges or chaining requests into unintended actions?
- Auditability – Can we trace every MCP call, understand who initiated it, and prove compliance during an audit?
- Runtime Observability – Do enterprises have real-time visibility into MCP interactions, or will they operate as black boxes?
- Governance Alignment – How does MCP map into existing frameworks like GDPR, HIPAA, PCI-DSS, or FedRAMP?
If these questions aren’t answered, enterprises are less likely to scale MCP beyond pilots and proofs of concept.
Early Signals: Emerging Solutions and Frameworks
The good news: awareness is growing. The Mirantis blog highlights the need for MCP-aware gateways and governance frameworks that echo what API gateways and service meshes achieved in the last decade.
A few patterns are already emerging:
- Secure Agent Gateways (e.g., AgentGateway, ArchGW) that sit between agents and MCP endpoints, enforcing policy and monitoring requests.
- Mesh Architectures inspired by McKinsey QuantumBlack’s vision of an Agentic AI Mesh, where agent interactions are brokered, observed, and governed as distributed systems.
- Enterprise Control Planes that extend existing identity and policy tooling into MCP contexts, mapping enterprise IAM, RBAC, and zero-trust models onto MCP traffic.
These early tools show promise, but they remain fragmented. What’s missing is a cohesive ecosystem where vendors, open source projects, and standards bodies work together to establish a secure foundation for MCP.
Analyst View: The Roadmap to Enterprise-Grade MCP
To cross the adoption chasm, MCP needs more than enthusiasm. It needs a security roadmap that enterprises can believe in. That roadmap likely includes:
- Security by Default
- Authentication and encryption must be embedded, not optional.
- Every MCP request should be verifiable, non-repudiable, and policy-bound.
- Governed Autonomy
- Agents should have bounded decision-making, with human-in-the-loop escalation paths.
- Just as Kubernetes enforces quotas and limits, MCP infrastructure should enforce guardrails on what agents can and cannot do.
- Observability at Scale
- Enterprises need dashboards, alerts, and forensic tools for MCP traffic.
- Without visibility, MCP becomes a blind spot in security operations.
- Compliance Alignment
- MCP protocols and gateways must demonstrate how they meet GDPR, HIPAA, SOC 2, and other regulatory frameworks.
- Enterprises won’t adopt en masse without documented compliance pathways.
- Ecosystem Maturity
- Vendors must collaborate on open standards for MCP governance.
- Proprietary lock-in will stifle trust; shared frameworks will accelerate adoption.
The Vendor Landscape: Who Wins?
The winners in this emerging space will not be those who build the flashiest MCP tools. Instead, they’ll be the vendors that:
- Integrate deeply with enterprise identity and security stacks (Okta, Microsoft Entra, HashiCorp Vault).
- Offer observability hooks that plug into SIEMs and SecOps workflows (Splunk, Dynatrace, Datadog, Elastic).
- Bridge the developer-enterprise gap, providing tooling that satisfies both innovation and compliance.
Mirantis, by raising the flag early, positions itself as a thought leader in this conversation. But expect security vendors, observability providers, and cloud platforms to quickly follow. The market opportunity is too large to ignore: the chance to become the API gateway equivalent for MCP.
Looking Ahead: MCP as the Digital Workforce Standard
The bigger picture is clear. If enterprises begin to treat AI agents as digital workers, MCP is their workstation. Securing that workstation is non-negotiable.
The parallels to cloud and API adoption are instructive:
- REST without OAuth would never have scaled.
- Cloud without IAM would have collapsed under regulatory pressure.
- MCP without governance risks the same fate.
If the AI ecosystem learns from history, MCP could accelerate into the enterprise mainstream. If it doesn’t, it risks being sidelined as another developer experiment too risky for production.
Security is the Price of Adoption
MCP has the potential to become the connective tissue of agentic AI. But connective tissue can also transmit infection. Unless MCP evolves into a secure, governable, and observable protocol, enterprises will hesitate to adopt it at scale.
The next 18 months are critical to adoption. Vendors who address these challenges head-on (secure gateways, policy frameworks, and compliance-ready infrastructure) will define the agentic AI ecosystem for the next decade.
As with REST and cloud before it, security is not optional for MCP. It’s the price of adoption.
 is critical for enterprise AI adoption and how governance, trust, and control will define its future.){kind=link}