MCP – connecting Claude to your own systems

In part 3, Claude used a skill to grab the top-priority ticket from our ticket system and create a branch for it. Which leaves one question: how does Claude get into the ticket system in the first place?

The answer is MCP, the Model Context Protocol.

🔌 What is MCP?

An open standard that lets AI applications talk to external systems. Think of it as USB-C for AI: an MCP server gives Claude tools such as "search tickets", "update ticket" or "log time". Claude decides on its own when it needs which tool.

What this means day to day:

🎫 Ticket system: Claude reads the requirements straight from the ticket instead of me pasting them in, and documents what was implemented at the end.
🗄️ Database: "Why is this record duplicated?" Claude looks it up instead of guessing. Ideally with read-only access to a test database.
📚 Docs and wiki: internal guidelines flow directly into the implementation.
🌐 Browser: Claude checks the UI it just built itself.

Setup:

Ready-made MCP servers exist for many systems, added with a command like claude mcp add. Put the configuration in the repository as .mcp.json, and the whole team shares the same connections.

And if there's no server for your system yet? A custom MCP server for an internal REST API is manageable, and Claude Code will happily write it for you.

A word of caution:

Every MCP server extends what Claude can do. So only use servers from trusted sources and grant permissions deliberately. More on that in the next part.

My rule of thumb:

• Claude needs to KNOW something or follow a PROCESS? Skill.
• Claude needs to talk to a SYSTEM? MCP.
• They're strongest together: the skill describes the workflow, MCP provides the access.

Which system would you connect to Claude first? Let me know in the comments

Join the discussion on LinkedIn →