Introduction
Before the Model Context Protocol, connecting AI models to external data sources required writing custom, point-to-point integration code for every combination of LLM and software platform.
In traditional agentic AI deployments, developers may build disparate, unmanaged connections to internal tools and databases, resulting in a fragmented approach. MCP, an open protocol introduced by Anthropic in 2024, provides a standardized approach with a framework that enables Large Language Models (LLMs) and AI agents to interact with external tools, APIs, and data sources through a standardized protocol.
While the Model Context Protocol (MCP) introduces much-needed standardization, unmanaged implementations introduce operational and security risks. Connecting numerous disparate tools without centralized oversight leads to tool bloat and integration complexity across the enterprise. Furthermore, MCP implementations can introduce excessive-agency risk when agents receive access to tools or permissions beyond what a task requires. Without appropriately scoped, function-level controls, over-privileged agents may be able to invoke actions beyond the requirements of a task or access sensitive internal data.
The Agent Gateway component of Nutanix Enterprise AI provides a centralized management layer for supported MCP server connections. Instead of maintaining separate integrations for every application, client applications connect through a single, aggregated /mcp route. Consolidating supported tool access through this gateway can help organizations centralize configured governance, auditing, and role-based access control (RBAC) for agentic workflows.
Tool Authorization & Credential Management
To support enterprise-defined security policies, administrators can use MCP Connectors and MCP Client Keys to configure:
- Tool-Level Filtering: MCP Connectors aggregate local or remote tool servers to configure tool exposure or access controls. Assigning connector configurations to approved client keys can help control which tool capabilities a client may be permitted to access or invoke.
- Credential Management: Storing connection credentials within the platform can reduce the need to place those credentials in client applications or agent code.
Reducing Tool Bloat
Tool-level filtering can help reduce the number of tools exposed to an agent and may reduce tool-schema context (sometimes referred to as tool bloat). When many tool definitions are available to an LLM, the additional tool schema may lead to unnecessary context, which can result in poor accuracy (context rot) and increased token consumption. Effects on model performance and hallucination rates vary by model, prompt, tool schema, and workload.
Deployment Options: Remote vs. Local MCP Servers
To support diverse hybrid environments, Nutanix Enterprise AI accommodates two deployment models:
- Remote MCP Servers: A remote MCP server is an externally hosted MCP server that Nutanix Enterprise AI connects to through a URL. Remote MCP servers are connected via supported HTTPS endpoints using Streamable HTTP transport and CA-signed TLS certificates, subject to applicable product and deployment requirements.
- Local MCP Servers: A containerized MCP server that is deployed and managed directly within the Nutanix Enterprise AI Kubernetes cluster. Nutanix Enterprise AI 2.8 supports container lifecycle management, including configured registry credentials, port mapping, and CPU and memory resource allocations.
Security Considerations for MCP Servers
To maintain cluster security and workload isolation, local MCP servers are deployed with strict controls on network access, runtime permissions, file system usage, and naming conventions. Egress is allowed to all destinations except the Kubernetes API server, and ingress is restricted to the nai-system and envoy-gateway-system namespaces. Local MCP servers cannot run as the root or any privileged user, and root filesystem access is read-only.
How It Works
- The administrator registers local or remote MCP servers.
- The administrator creates an MCP connector to aggregate one or more servers.
- The administrator associates MCP Client Keys with the connector and configures the tools exposed through that association, and which tools from each server the associated key is allowed to access.
- A client connects to the MCP connector using its configured client key and can access the supported servers and tools exposed through that connector configuration.
Securing the Future of Agentic Workloads
Nutanix Enterprise AI provides a centralized management and access-control layer for supported MCP connections, helping organizations manage integration sprawl and apply configured controls to agentic workflows. Consolidating tool access behind a single governed gateway interface (/mcp) can help agents access approved internal tools and data sources while helping reduce integration complexity. With supported tool-exposure or authorization controls, credential-management capabilities, and local or remote deployment options, organizations can help manage risks to corporate data assets while supporting configured auditing and access-control workflows.
To see for yourself how to configure MCP connectors and servers, and other Nutanix Agent Gateway capabilities, take the Nutanix Enterprise AI Test Drive and select the Agent Gateway experience.
©2026 Nutanix, Inc. All rights reserved. Nutanix, the Nutanix logo and all Nutanix product and service names mentioned are registered trademarks or trademarks of Nutanix, Inc. in the United States and other countries. All other brand names mentioned are for identification purposes only and may be the trademarks of their respective holder(s). Our decision to publish, link to, or reference third-party data or other information should not be considered an endorsement of any such data or information. If you decide to leave our content and access these third-party studies, publications, surveys, and other data/information, then you do so at your own risk.