Cockpit MCP: connecting your business software to AI in 2026

At Fragments Studio, we use Cockpit to run all of our projects and to give our clients a centralized collaboration space: contracts, invoices, working documents, meetings, project tracking… everything is one click away. By building our own MCP server, we turned this collaborative platform into a tool that can interact natively with AI agents, paving the way for a new era of intelligent automation for business software.

2025 marked a turning point with the mass adoption of autonomous agents. In 2026, the question is no longer whether AI can write an email, but whether it can directly manipulate your business data to carry out complex tasks. That is precisely why we equipped Cockpit, our project management platform, with an MCP (Model Context Protocol) server.

What is Fragments Cockpit, and why open it up to AI?

Fragments Cockpit is our agency's operational hub, but also our main collaboration space with our clients. It is custom software that centralizes project management, time tracking and invoicing, while giving each client direct, secure access to all of their contracts, invoices, working documents, meeting notes, meetings and the real-time progress of their projects.

In practice, our clients no longer need to dig through their emails or chase a project manager to find a signed quote or the week's schedule: everything is centralized in Cockpit.

With the MCP server, this collaboration goes one step further: our clients can now ask their favorite LLM, Claude, ChatGPT, Gemini or any other compatible assistant, and instantly access all of Cockpit's information in natural language, without even opening the interface.

Until now, despite this rich functionality, Cockpit remained a passive tool: a human had to read the information and act on it.

The emergence of the Model Context Protocol (MCP), a standard initiated by Anthropic, changed the game. An MCP server is a standardized interface that lets a large language model (LLM) such as Claude, GPT-4 or Gemini securely access local or remote data and execute specific functions (called "tools").

By building this bridge, we allow our AI tools to "read" the state of a project in Cockpit, "understand" who is working on what, instantly find a contract or an invoice, and "write" reports or update tasks without any manual intervention.

According to a McKinsey study published in late 2025, integrating AI agents into operational workflows can reduce the time spent on administrative tasks by 35% to 45%.

The technical project: how we built Cockpit's MCP server

The development of Cockpit's MCP server followed a rigorous architecture to guarantee security and performance.

We chose an implementation in TypeScript using the official Model Context Protocol SDK.

1. Defining resources and tools

The first step was to expose our data schemas as resources. In the MCP standard, a resource is static data that the AI can consult (e.g. the client list, the details of a quote, ongoing contracts or invoicing history).

We then defined tools, which are functions the AI can execute:

• get_project_status(projectId): to retrieve the actual progress.
• log_time(userId, duration, description): to record hours worked.
• create_task(projectId, title, deadline): to schedule a new step.

2. Security and authentication via API key

Security is the critical point. The MCP server acts as a gateway.

We implemented a robust authentication layer in which each AI agent uses a specific API key tied to granular permissions. The AI can never access data that a standard user would not be allowed to see.

At Fragments Studio, we apply the principle of least privilege: the reporting agent only has read access, while the planning agent can write to the tasks module.

3. Transport and interconnection

We configured the server to support transport via Stdio (for local use on our development machines) and via HTTP/SSE (Server-Sent Events) for our cloud deployments on AWS.

This flexibility lets our developers use Cockpit directly from their IDE (such as Cursor or VS Code) while allowing our global agents to access it over the web.

Why is MCP a revolution for your business software?

If your company uses internal management software (ERP, CRM or a niche tool), you are probably suffering from "siloed data" syndrome. Your data is there, but it is hard to use without a human to connect the dots.

Here is why building an MCP server for your custom tool is a strategic investment in 2026:

• Universal interoperability: Instead of coding a specific integration for each new AI tool, MCP provides a common language. Once your MCP server is deployed, any compatible agent can communicate with your software.
• Less context-switching: Your team members no longer waste time navigating between 15 tabs. They simply ask their assistant: "What is the remaining budget on project X?" and the AI queries Cockpit instantly via MCP.
• Intelligent automation: Unlike traditional automations (such as Zapier) which are linear, agents using MCP can make decisions based on the complex context of your business data.

We have seen with our clients that a business tool equipped with an AI interface sees its internal adoption rate jump by 60%, because access to information becomes conversational and instant.

Toward "Agent-Ready" companies

Implementing the MCP server on Cockpit is only the first step.

We are now considering workflows in which AI agents assign themselves minor maintenance tickets based on the workload detected in the tool, or generate proactive alerts when a project is at risk of exceeding its budget.

For an SMB, turning its internal software into an Agent-Ready platform is no longer a luxury option, it is a condition for competitiveness. It frees human talent from data entry and search tasks so they can focus on decision-making and creativity.

Frequently asked questions

What is an MCP server, concretely? It is a small computer program that acts as an interpreter between your database and an Artificial Intelligence. It exposes your data in a way that a model like Claude can understand and interact with securely.

Do I need to rebuild my entire software to add MCP? No. An MCP server can be added as an extra layer on top of your existing API. It is an extension that turns your standard endpoints into tools an LLM can understand.

Is my data safe with MCP? Yes, provided the server is implemented properly. The MCP standard does not give free access to your database. You define precisely which "tools" and which "resources" are exposed to the AI, with full control over authentication.

What is the real productivity gain? By eliminating the manual back-and-forth to look up information or update a status, we generally see time savings of 15 to 20 hours per month per team member on administrative and project management functions.

Conclusion

By opening Cockpit to the Model Context Protocol, we did not just add a feature: we changed the very nature of our collaboration tool. From a platform where clients and teams consulted contracts, invoices and working documents, Cockpit has become an active member of our organization, able to feed AI in real time and automate the interactions that used to waste time for both sides.

If you have internal business software, it is time to ask yourself how to make it communicate with the agents that now dominate the 2026 technology landscape.

Prepare your infrastructure for the era of AI agents

Moving to the Model Context Protocol is a major step toward getting more value from your proprietary data and improving operational efficiency.

• Discover our AI development expertise to transform your internal tools.
• If you want to modernize your infrastructure, our custom web development service supports you in implementing standards like MCP.
• Get an initial estimate for your interconnection project with our project estimator.
• Or let's talk directly to assess whether your current software is eligible.