MCP (Model Context Protocol)
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external tools and data sources through a single, consistent interface. Introduced by Anthropic in November 2024, it means one AI agent can read from and act inside systems such as a CRM, calendar, or database without a bespoke integration for each one.
What problem does MCP solve for businesses running multiple systems?
Before MCP, connecting an AI agent to external systems meant writing a custom implementation for every new data source. A team running HubSpot, Xero, and Outlook alongside an AI assistant would need three separate connectors, each maintained individually. Anthropic designed MCP to address this fragmentation directly: build an MCP server for a system once, and any MCP-compatible AI client can connect to it. The result is a single, reusable integration layer rather than a growing backlog of point-to-point connections.
How does MCP actually work?
MCP follows a client-server architecture with three roles. The host is the AI application (such as Claude or an AI-powered CRM assistant). The client is the connection manager inside the host. The server is a programme that wraps a specific tool or data source and exposes its capabilities through the standardised protocol. Servers advertise what they can do in natural language; the AI selects the right tool and sends a structured request. Results are returned in a consistent format. Communication runs over JSON-RPC 2.0, making it language-agnostic and straightforward to implement in any standard tech stack.
Which AI platforms and companies support MCP?
MCP launched in November 2024 as an Anthropic open-source project and reached industry-standard status within a year. OpenAI integrated MCP into ChatGPT and its Agents SDK in March 2025. Google DeepMind confirmed support in April 2025. Microsoft integrated it with Azure OpenAI and Semantic Kernel. In December 2025, Anthropic donated governance of the protocol to the Agentic AI Foundation (AAIF), a directed fund under the Linux Foundation, co-founded by Anthropic, Block, and OpenAI, with Google, Microsoft, AWS, Cloudflare, and Bloomberg as supporters. SDK downloads reached 97 million per month by March 2026, up from roughly 100,000 at launch.
How does MCP relate to APIs and other integration approaches?
An API defines what a specific system can do and how to call it. MCP sits one layer above: it provides a standard for how an AI agent discovers and invokes any API or tool, regardless of which system sits underneath. Where a traditional API integration is point-to-point, an MCP server acts as a bridge that an AI can find and use without the developer pre-wiring every connection. For operators already running several software platforms, this means future AI tooling can be wired to existing systems through a single, reusable layer rather than rebuilding integrations with each new capability added.
Key takeaways
- MCP is an open standard, now governed by the Linux Foundation, that gives AI agents a single, consistent way to connect to external tools and data sources.
- Any MCP-compatible AI client can use an MCP server, regardless of the underlying AI model or platform: OpenAI, Google DeepMind, and Microsoft have all adopted the protocol.
- MCP sits above individual APIs. You build a server for a system once; every AI tool you add connects to the same server rather than requiring new point-to-point integrations.
- Access is scoped by design: each MCP server exposes only the capabilities you define, so sensitive systems are not exposed beyond what is needed.
- As of March 2026, the MCP SDK is downloaded 97 million times per month, reflecting rapid adoption across the AI industry.
How Cloudfox Helps With MCP
Cloudfox builds and operates MCP-connected stacks for PBSA and BTR operators. Our own Chief of Staff AI agent runs on MCP, connecting to HubSpot (contacts, deals, tasks), Outlook (email and calendar), Xero (financial data), and internal databases through a single protocol layer. When we implement HubSpot for an operator, we design the integration architecture so that AI tooling can be added without rebuilding connections from scratch each time: MCP servers wrap your CRM, your finance stack, and your property management system, and any AI agent you introduce can address all three through one standard interface. The result is a system that gets more capable as AI matures, without generating a maintenance backlog of bespoke connectors. To see how this fits our broader implementation approach, visit cloudfox.it/what-we-do.
Frequently Asked Questions About MCP
Do we need MCP if we already have API integrations?
Existing API integrations keep working. MCP sits above them: it gives any AI agent a standard way to discover and call those integrations without you writing additional glue code per agent. If you add a second AI tool in future, it connects to the same MCP servers rather than requiring a fresh round of custom API work.
Is MCP only for Anthropic's Claude, or does it work with other AI tools?
MCP is an open standard now governed by the Agentic AI Foundation under the Linux Foundation. OpenAI, Google DeepMind, and Microsoft have all adopted it. An MCP server you build today works with any MCP-compatible AI client, regardless of the underlying model.
How does MCP differ from a standard CRM integration or Zapier workflow?
Zapier and point-to-point CRM integrations move data between systems on a fixed trigger-action basis. MCP lets an AI agent reason about which tool to use and when, then call it dynamically. The agent decides the sequence; it is not pre-scripted. This supports more complex, multi-step tasks that would require separate automations to cover every scenario.
What does MCP mean in practice for a PBSA operator?
In practical terms, MCP is what allows an AI assistant to check a HubSpot deal, pull the related invoices from Xero, and draft a response in Outlook in a single step, without a human switching between three systems. The AI connects to all three through MCP servers and handles the retrieval and action in sequence.
Is MCP secure to use with sensitive property management or financial data?
MCP servers are built and hosted by the operator or their implementation partner, not by Anthropic or the AI provider. Data stays within your chosen infrastructure. Each server exposes only the capabilities you define, so access is scoped by design. As with any integration, the security posture depends on how the server is built and where it is hosted.