Most conversations about AI agents focus on what agents can do. The demos are impressive. An agent that can research a prospect, draft an outreach email, and log the activity to your CRM without a human touching it at any step is genuinely useful. I see businesses running workflows like this every week.

What rarely gets discussed in those conversations is the layer underneath all of it. The plumbing that makes the agent capable of connecting to your CRM in the first place, or your calendar, or your internal reporting system, or the custom database that your operations run on.

That plumbing problem is exactly what the three best-funded new startups in enterprise AI are trying to solve. And understanding it will change how you think about deploying agents in your own business.

What MCP is and why it matters

Anthropic published the Model Context Protocol in November 2024. By March 2026 it had reached 97 million monthly SDK downloads. OpenAI, Google DeepMind, Microsoft, and AWS all adopted it. Anthropic donated it to the Linux Foundation, co-founding the Agentic AI Foundation alongside OpenAI and Block.

The protocol solves a problem that sounds simple but is genuinely painful in practice. An AI agent needs to connect to external tools and data. Before MCP, every connection was a custom build. If you wanted your agent to read from your CRM and write to your project management tool, someone had to build that integration from scratch, maintain it, and rebuild it every time either system changed.

MCP is a standardised way for agents to connect to any system that has an MCP server. The analogy I use is HTTP. Before HTTP, every website needed its own custom protocol for browsers to talk to it. After HTTP, any browser could talk to any website that spoke the same standard. MCP does the same thing for AI agents and the tools they connect to.

For businesses thinking about deploying agents, this matters because it dramatically lowers the cost of connecting your agents to your existing systems. You are not commissioning bespoke integrations for each tool your business runs on. You are using a standard protocol that an expanding ecosystem of tools is adopting.

The three companies that got funded to solve the plumbing

Three seed-stage companies raised capital in the past twelve months to build the infrastructure layer above MCP. Together they tell you something about where the enterprise AI stack is heading.

Manufact raised $6.3 million in February 2026, led by Peak XV with Y Combinator, Liquid 2 Ventures, and Ritual Capital. The company was in the YC S25 batch under its original name, mcp-use, before rebranding. It builds an open-source SDK and cloud hosting layer for MCP servers. The pitch is essentially what Vercel did for Next.js: take a protocol that is powerful but technically demanding to run in production, and make it deployable by teams that are not infrastructure engineers.

Runlayer raised $11 million in November 2025, led by Khosla Ventures via Keith Rabois and Felicis. Runlayer addresses the security problem specifically. When your AI agents connect to real business systems — your CRM, your financial data, your internal databases — those connections need proper access controls, audit logs, and threat monitoring. Traditional enterprise security tools were not designed for this use case. Runlayer builds the security layer that enterprises need to run MCP deployments safely.

Alpic raised $6 million in pre-seed funding in September 2025, led by Partech. Alpic is building what it calls an MCP-native cloud platform: infrastructure designed from the ground up for the operational characteristics of MCP servers, rather than adapting general-purpose cloud infrastructure that was designed for human-driven software.

Three companies, three different layers of the same stack. Peak XV and Khosla are not firms that make early bets in thin markets. The fact that they are in this space tells you that serious investors believe the MCP infrastructure layer is going to be a significant market.

Why the plumbing problem is actually the agent adoption problem

When I work with businesses on agent deployments through Omni Ops, the most common point of friction is not the agent itself. Modern AI systems are capable enough for a wide range of business automation tasks. The friction is getting the agent connected to the systems the business actually runs on.

Every business has a different stack. Some use Salesforce. Some use a custom-built CRM. Some use a mix of tools that have been accumulating for a decade. Some have databases that were built before most cloud services existed. The agent needs to be able to read from and write to whatever combination of systems that specific business uses.

Without a connectivity standard, every deployment is a custom integration project. That adds time, adds cost, and creates ongoing maintenance obligations. It also means that when a system changes its API, or when you add a new tool to your stack, someone has to go back and update the agent’s connections.

With MCP becoming an industry standard, the picture changes. Systems that have MCP servers can be connected to by any MCP-compatible agent, with the authentication and authorisation handled by infrastructure like Runlayer rather than built custom each time. The cost and time of connecting an agent to your business systems drops significantly.

That is why I pay close attention to which tools in the business software market are shipping MCP support. Every tool that ships an MCP server is a tool your AI agents can connect to without a custom integration project.

What this means for how you deploy agents

The practical implication for businesses thinking about agent deployment is to stop treating connectivity as a later problem.

The question of what systems your agents will connect to is not a technical detail to sort out after you have decided to deploy. It is the question that determines how expensive and how fast deployment will be. Starting with a clear picture of which systems your agents need to access, which of those have MCP support, and what the security requirements are for those connections is the work that makes deployment go smoothly rather than slowly.

The businesses that are deploying agents fastest right now are the ones that have thought through this connectivity layer. They are not waiting until they have a fully built agent to figure out how it talks to their data. They are mapping that out at the design stage.

The MCP ecosystem funding is telling you that the broader industry is solving the plumbing problem at scale. Manufact, Runlayer, and Alpic are building the infrastructure that will make agent connectivity reliable, secure, and operationally manageable for enterprises.

The question for your business is whether you are waiting for that infrastructure to mature further before starting, or whether you are building now and using the emerging standards to reduce your integration costs as they develop.

In my experience, the businesses that wait for perfect conditions are always catching up to the businesses that started moving when the conditions were good enough. The MCP ecosystem is good enough now. The tools that spoke the protocol six months ago are still there. The ecosystem has only grown.

The compounding effect of good infrastructure choices

When you make a good infrastructure choice early, you accumulate advantages. The data your agents generate becomes more structured. The integrations you build become easier to maintain. The processes you automate become composable building blocks for the next automation.

When you make a bad infrastructure choice early, you accumulate debt. Custom integrations that break. Workarounds that every new hire has to learn. Agent deployments that cannot be extended without going back to the beginning.

MCP is the right infrastructure choice for agent connectivity. The funding behind Manufact, Runlayer, and Alpic, and the adoption by every major AI and cloud provider, is strong evidence of that. Building your agent deployments on this foundation now is building on infrastructure that will be more capable, better supported, and more widely integrated in twelve months than it is today.

---

Through Omni Ops, we design and deploy AI agent workforces that connect properly to the tools and data your business already uses. If you want to understand what that looks like for your specific stack, book a session.