A standard that happened faster than anyone expected
In November 2024, Anthropic released the Model Context Protocol (MCP) as an open standard. At the time it was a technical proposal: a common way for AI models to connect to external tools and data sources, replacing the mess of one-off integrations that had plagued early AI deployments. Few predicted how quickly it would take over.
By March 2026, the picture looks very different. In December 2025, Anthropic donated MCP to the Linux Foundation's new Agentic AI Foundation, co-founded with Block and OpenAI, with Google, Microsoft, Amazon Web Services, and Cloudflare also on board. Today there are more than 10,000 active public MCP servers. ChatGPT, Google Gemini, Microsoft Copilot, Cursor, and Visual Studio Code all support the protocol natively. Downloads run at seven million per month.
San Francisco-based startup Manufact, which emerged from Y Combinator's Summer 2025 batch, just raised $6.3 million in seed funding to build open-source tooling and cloud infrastructure for MCP. Lead investor Peak XV (formerly Sequoia Capital India and Southeast Asia) called MCP 'the USB-C for AI.' Their thesis: software that does not offer an MCP interface will soon be invisible to AI agents and the users who increasingly rely on them.
Why this matters for enterprise software
Before MCP, connecting an AI agent to a company's internal systems meant custom integration work for every single tool. A bespoke connector for your ERP, another for your CRM, another for your document management system. Each one was expensive to build and fragile to maintain. MCP changes the equation: one protocol, consistent across models and platforms, that lets any AI agent communicate with any system that exposes an MCP server.
The implications are significant. The global AI agents market reached $7.84 billion in 2025 and is projected to hit $52.62 billion by 2030. That growth is built on agents that can actually do things: process invoices, query order histories, update customer records, generate reports from live data. None of that works without the connective tissue between the AI and the systems that hold your business data.
Manufact's co-CEO Luigi Pederzani draws a direct comparison to the mobile transition: 'Who would have bought a hotel or a flight from a mobile app in the early days? But then the web became mobile first.' His prediction: software products will become MCP first. Businesses that wait to build proper AI integration infrastructure will face the same problem as companies that had no mobile strategy in 2015.
What this means in practice
At Laava, we have been building AI integrations for Dutch businesses for two years. The emergence of MCP as a universal standard changes the economics of that work significantly, but it does not make it simple. The protocol defines how AI agents communicate with external systems. It does not define what those systems expose, how access is secured, or what the agent is actually allowed to do with the data it retrieves.
When we build an AI agent that processes incoming invoices and pushes extracted data to an ERP like Exact Online or SAP, the MCP layer is one part of a larger architecture. You still need to define which data the agent can read and write, build validation logic, handle exceptions, and ensure that humans stay in the loop for decisions that carry financial or legal weight. Getting the integration right is where the real work sits.
The good news is that with MCP now a true standard, the integration work Laava builds is more portable than ever. An MCP server we build for one client's document system can connect to Claude today and to whatever model performs best in two years, without rebuilding the connector. The investment in proper AI integration infrastructure is no longer locked to a specific model or vendor.
Where to start
If you are evaluating AI for your organisation, the first question is not which model to use. It is which systems hold the data your AI needs to be useful, and what it would take to connect them properly. Document flows, ERP data, customer communication histories: these are the sources that determine whether an AI agent can actually automate meaningful work or just answer generic questions.
Laava runs four-week pilots that start with exactly that question. We map your highest-value document or workflow process, build the integration layer including MCP servers where appropriate, and deliver a working automation you can evaluate against real business output. If you want to understand what proper AI integration looks like for your systems, reach out.