Digital Realty brings MCP to ServiceFabric network layer

Digital Realty, one of the world's largest data centre operators with over 300 facilities across 25 countries, has announced native support for the Model Context Protocol (MCP) on its ServiceFabric platform. The move transforms an interconnection layer designed for software-defined networks into an environment directly accessible by AI agents, according to SDxCentral reporting on 19 June 2026.

This is no minor announcement: that a physical infrastructure company adopts a protocol designed for LLMs to invoke external tools signals how far MCP has travelled as a de facto standard beyond the development environment.

What ServiceFabric is and what changes with MCP

ServiceFabric is Digital Realty's programmable interconnection platform. It allows its customers (cloud operators, enterprises with multi-datacenter presence, telecom providers) to configure and manage connectivity between physical facilities through a software layer. Until now, that programmability was exercised through conventional APIs: a human operator or an automation system would issue REST calls to provision circuits, adjust bandwidth or monitor latency.

With MCP, ServiceFabric exposes its capabilities as an MCP server that an AI agent can connect to directly. In practice, this means an agent running on Claude or any other protocol-compatible client can query network status, request configuration changes or respond to infrastructure events without a human intermediary translating intentions into API calls.

The protocol, developed by Anthropic and now managed as an open standard, defines how an LLM describes available tools, invokes them and receives results. Digital Realty implementing it in ServiceFabric is not technically complex (it essentially wraps existing APIs in an MCP server), but the gesture carries weight: it normalises the idea that network infrastructure should be readable and operable by agents, not just by humans or scripts.

Why this matters now

The context is not incidental. The expansion of distributed computing for AI inference has put data centre operators in an awkward position: their customers need to provision resources at a speed and granularity that manual processes cannot keep up with. GPU capacity expansion cycles, the need to redirect traffic between regions based on inference load or dynamic SLA agreements are scenarios where an agent with direct access to the network layer has clear advantages over an operator working with dashboards.

Moreover, Digital Realty competes in a market where Equinix, CoreSite and other major operators are also making moves towards intelligent network automation. Offering an "MCP-native" environment is a way to differentiate with customers who are already building agent pipelines on Claude Code or other compatible platforms.

Who benefits immediately

The profiles that can benefit from this integration right now are relatively narrow, but real:

• Infrastructure teams at enterprises with multi-datacenter presence already experimenting with network automation agents.
• Managed service providers (MSPs) managing connectivity for multiple customers who want to delegate routine monitoring or adjustment tasks to specialised sub-agents.
• MLOps platforms needing to provision or release network capacity based on training or inference load, without manual intervention in each cycle.

For the individual developer or small team without access to ServiceFabric, the announcement is more of a signal: MCP is ceasing to be a protocol for connecting LLMs with productivity tools and is entering the layer of enterprise infrastructure.

A note on protocol status

MCP has been active since late 2024 and in recent months we have seen a notable acceleration in the number of organisations publishing their own MCP servers. Most remain SaaS integrations (task managers, databases, third-party APIs). That a physical infrastructure operator at Digital Realty's scale appears on this list is a qualitative jump, regardless of the technical implementation details.

From our perspective, we value positively that the MCP ecosystem is gaining traction in deeper layers of the stack. That said, it will be important to see with what maturity and what security controls this integration reaches production: giving an AI agent write access to critical network infrastructure is exactly the type of case where authorisation and audit details cannot take a back seat.