Workshops: Agentic Coding

Building MCP Servers: Hands-On Workshop for Dev Teams

The Model Context Protocol (MCP) is an open standard that lets AI assistants and agents access tools and data. In this workshop your team builds its own MCP server for an internal system and learns to make it secure, maintainable and usable across multiple AI clients.

Who is this workshop for?

  • Backend and integration developers
  • Platform and developer experience teams
  • AI engineers and AI product teams
  • Software architects
  • Teams opening internal tools to AI agents

What you will take away

  • Understand the MCP architecture of hosts, clients and servers and the building blocks of tools, resources and prompts.
  • Build a working MCP server with an official SDK, for example in TypeScript or Python.
  • Design tool descriptions and schemas so models use them correctly and reliably.
  • Implement authentication, permissions and protection against prompt injection.
  • Test, deploy and run MCP servers locally and remotely.

Agenda

  1. Module 1

    Understanding MCP

    • Architecture: host, client, server
    • Tools, resources and prompts
    • Transports: local via stdio and remote via HTTP
  2. Module 2

    Your first server

    • Exercise: MCP server for an internal system using the SDK
    • Testing with the MCP Inspector
    • Connecting to Claude Code, Claude Desktop, Cursor or GitHub Copilot
  3. Module 3

    Good tool design

    • Names, descriptions and input schemas for models
    • Granularity: a few powerful tools vs. many small ones
    • Error messages that help the model recover
  4. Module 4

    Security

    • Authentication and OAuth for remote servers
    • Least privilege, read vs. write tools, confirmations
    • Prompt injection via tool results and tool poisoning
  5. Module 5

    Operations and distribution

    • Deployment, logging and monitoring
    • Versioning and compatibility
    • An approval process for MCP servers in the organisation

Why build your own MCP servers?

MCP servers already exist for many standard services. The real value for organisations comes when agents can reach internal systems: ticketing, product databases, internal APIs or knowledge bases.

A well-built MCP server works with multiple clients. You build the integration once and use it in Claude, in coding agents and in other MCP-capable tools.

Security is not an afterthought

MCP servers give AI systems the power to act. So we treat security as a thread running through the whole day, not an appendix.

  • Expose only the tools and rights needed
  • Confirmations for write actions
  • Validate inputs, limit outputs
  • Vet third-party MCP servers before use

Format

An intensive hands-on day with neooutput AI Code Engineers. Remote or on-site, in English or German. You leave with a running server for your use case.

Frequently asked questions

Which programming language do we need?

There are official SDKs for several languages, including TypeScript, Python, Java, Kotlin, C# and Go. We tailor the workshop to your preferred language.

Does MCP only work with Claude?

No. MCP was introduced by Anthropic as an open standard and is now supported by many clients, including ChatGPT, GitHub Copilot, Cursor and Windsurf.

Should we build local or remote servers?

Local servers suit developer tooling; remote servers suit integrations many users share centrally. We discuss what fits your case.

Workshop

Get in touch

Which internal system should your AI agents be able to use first?

Tell us a little about your team – we will get back to you within 24 hours with a proposal for content, format and dates.

Prefer to talk?

We only use your details to handle this request. See our privacy policy.