Skip to content

What is MCP, and how to add a server to Claude Code

2 min read

A coding agent that can only read your files and run commands is useful. One that can also query your issue tracker, read your design files or look at a database is more useful. MCP is the standard way to give it those abilities.

What MCP is

MCP stands for Model Context Protocol, an open standard for connecting AI applications to tools and data. Instead of every app inventing its own plug-in format, a tool is written once as an MCP server, and any app that speaks the protocol can use it.

Three terms cover most of what you need:

  • Client: the app that talks to the model, such as Claude Code.
  • Server: a small program that exposes tools or data, such as "search issues" or "read this file".
  • Tool: one action a server offers. The agent decides when to call it.

When you ask the agent something, it can see which tools its connected servers offer and call them when they help, much as it already runs shell commands.

Adding a server to Claude Code

Claude Code manages servers from the command line. The exact flags can change between versions, so check claude mcp --help for your install. The usual shape is:

# a server that runs as a local command
claude mcp add my-server -- npx -y some-mcp-package

# a server that runs elsewhere and is reached over HTTP
claude mcp add --transport http my-remote-server https://example.com/mcp

Replace the names and addresses with the ones from the server's own documentation. Then confirm it was registered:

claude mcp list

Inside a running Claude Code session, the /mcp command shows the status of your servers, and is also where servers that need you to sign in will ask for it.

Choosing where a server applies

You can usually register a server for yourself everywhere, for one project, or for one project and your own machine only. A server your whole team needs belongs in the project so everyone gets it; a server with your personal credentials should stay on your machine. Check the current options in the Claude Code documentation before relying on a specific flag.

Staying safe

An MCP server runs code and can reach real data, so treat adding one like installing software:

  1. Only add servers you trust. Prefer ones from the service's own vendor or a source you can read.
  2. Give it the least access it needs. A read-only token is better than an admin token.
  3. Watch the permission prompts. Claude Code asks before using tools it has not been allowed to use. Read the prompt rather than approving by reflex.
  4. Remove what you stop using. Unused servers are unused risk.

Working with MCP in Tidepane

Claude Code runs in a terminal, and Tidepane is built to host terminals running coding agents. If an agent stops to ask permission for an MCP tool, the terminal's status in the sidebar shows that it needs input, so you can find it even when several agents are running across projects.

Where to go next

Start with one server for something you already do by hand, such as reading tickets or checking a staging database. Once you see the agent use it well, add a second. Keep the list short, and you will always know what your agent can reach.

Tidepane brings terminals, a browser and your coding agents into one window. See what it does.