AI Agents and MCP Servers for Web Projects
What I Mean by AI Agents
An AI agent is a language model that does more than answer. When needed, it calls functions and keeps working with the result until the task is done. The model does not run the function itself: it returns a structured call, and your application runs it. That is where I come in. I build the functions an agent may call and the limits it works within.
The offer is aimed at small and medium-sized businesses (SMEs) in the DACH region with an existing website, an existing shop or an existing web application. For an overview of all AI services, see AI Integration.
What an Agent Can Do in Your Application
The agent gets a small selection of narrowly defined functions from your application. Examples:
- Look up an order status or free appointments and phrase the answer clearly.
- Create a draft that an employee reviews before it counts.
- Assign an incoming request to the right place in your application.
Where an agent takes over work and where conventional programming is the better choice, we work out together before I build anything. On every request, the model itself decides whether and which function to call. That is why I keep the number of functions small and describe each one precisely.
What an MCP Server Is
MCP stands for Model Context Protocol. It is a standard for connecting AI applications to external systems, for example data sources and tools. An MCP server provides an AI application with such tools, which it can call. Claude and the OpenAI API can connect to MCP servers.
For your project this means: I make selected functions of your web application reachable through this standard and define which data and actions belong to it.
This is how a call works:
- The AI application asks the server which tools it offers.
- The model picks a tool based on the request.
- The server runs the function and returns the result.
- The model uses the result for its answer.
Control and Limits
- Approval: The MCP specification recommends that a person should always be able to step in and deny tool calls. OpenAI, too, asks for approval by default before data goes to an MCP server. For actions with consequences I build such an approval in.
- Trust: Only connect MCP servers you trust. A manipulated server can siphon off data that reaches the model’s context. I connect third-party servers only after agreeing it with you.
- Access: For servers I build for you, I check inputs, assign access rights and limit the number of calls.
- Expectations: I do not promise a percentage. An agent remains a tool that you control.
What I Do Not Offer
I rely on existing AI services, meaning API and cloud providers, including ones in the EU. An analysis of your internal workflows and general advice on AI direction are not part of my offer. My technical advice on provider, model, location and costs serves the implementation and ends with it. It is not legal advice.
Discuss an Agent Project
Do you have an application in mind that an agent or an MCP server could sensibly extend? Briefly describe your system and your plan, and I will tell you whether and how it can be built.
Related Services
- AI Integration - all AI services at a glance
- AI Chatbot Development - answers from your own content
- AI Document Processing - read documents, review results in an editor
- AI and Hosting in the EU - location, model choice and third-party services
- AI Developer for Agencies - working together as a subcontractor
Frequently Asked Questions
What is the difference between an AI agent and a chatbot?
A plain chatbot answers with text. An AI agent can also call functions that I define in your application and keeps working with the result. If visitors mainly ask questions about your content, a chatbot is often enough, see AI chatbot development.
Do I need an MCP server for an AI agent?
Not necessarily. An MCP server makes sense when selected functions of your application should be reachable by AI applications through a shared standard. For an agent inside your own web application, a direct connection can be enough. We decide together after a look at your system.
Can an agent change or delete data on its own?
Only what I explicitly allow it to do. For actions with consequences, such as writing, deleting or sending, I plan in an approval by a person. The AI produces a suggestion, the decision stays with you.
Where does the data run?
We decide the location together before I build anything. I rely on existing AI services, meaning API and cloud providers, including ones in the EU. In the technical implementation I take GDPR requirements into account; whether a specific use is lawful is assessed by your data protection advisor. More under AI and hosting in the EU.
What does an AI agent or MCP server cost?
That cannot be answered in general. The price depends on how many functions the agent may call, how well your existing system can be connected, which permissions and approvals are needed and what usage costs the AI service causes. After a short conversation I can give you a concrete estimate.