← Claude Code, start to finish
Claude Code

MCP — How Claude Code Connects to Your Real Tools

30 Jul 2026·6 min read

With this post we step into the Advanced stage of the roadmap. Today we cover the feature that opens the most doors: MCP, the Model Context Protocol. By default Claude Code sees your code and runs commands, but it cannot see your Jira board or your database. MCP changes that. All of it is checked against the official docs.

What MCP is

MCP stands for Model Context Protocol. The docs call it an open source standard for AI-tool integrations. In simple words, it is one common language that lets Claude Code talk to outside tools, databases, and APIs. Instead of you copying data from another tool into the chat, Claude reads and acts on that system directly.

Why it is useful

Connect a server when you notice yourself copying data from another tool, like an issue tracker or a dashboard. Then you can say things like add the feature from Jira ENG-4521 and open a PR on GitHub, or check Sentry for how a feature is used, or pull user emails from our PostgreSQL database. Claude works from the live system, not from what you paste.

Add a server in one line

# hosted server over HTTP
claude mcp add --transport http claude-code-docs https://code.claude.com/docs/mcp

# a local tool as a subprocess (stdio)
claude mcp add playwright -- npx -y @playwright/mcp@latest

# check the connection status
claude mcp list
Add a hosted server with a URL, or a local tool after the -- separator.

When you add a server it is saved at a scope. Local is the default and stays private to you in this project. Use --scope user to make it work in all your projects, or --scope project to share it with your team through a .mcp.json file in the project root. That .mcp.json is worth committing to git, so everyone gets the same servers.

The /mcp command

Inside a session, /mcp is your control panel. It lists your servers and their status, lets you sign in to the ones that need a browser login, and lets you reconnect if a server drops. From the shell, claude mcp list gives the same status check before you even open a session.

The big ideaMCP is the bridge between Claude Code and the rest of your stack. One open standard, one add command, and a scope that decides who shares it. Start with one tool you copy data from every day and let Claude work from the source.

Try it: connect your first MCP server

Enough theory — let us connect one real server from start to finish. We will use the Claude Code docs server from the official quickstart, because it is hosted, needs no sign-in, and is safe to test with. The same three steps work for any server: add it, check it, then use it. Run the add and list commands in your terminal, not inside a claude session.

  • Step 1 — Add the server. In your terminal run the add command below. Claude Code prints a confirmation like Added HTTP MCP server claude-code-docs with URL: https://code.claude.com/docs/mcp to local config, which means the server is registered for you in this project.
  • Step 2 — Check the status. Run claude mcp list. The docs server should show as connected, which means it is ready to use. A local stdio server can briefly show a failed status while npx downloads the package, so wait a moment and run it again.
  • Step 3 — Use it in a session. Start claude and ask it to use the server by name, for example: Use the claude-code-docs server to look up what MCP_TIMEOUT does. The first time Claude calls the server it asks for permission — approve it, and the tool call in the output is labeled with the server name.
  • Step 4 — Sign in when needed. Some servers like Sentry show Needs authentication instead. For those, open a session, run /mcp, select the server, and choose Authenticate — your browser opens for sign-in, then the status turns to connected.
# Step 1: add the hosted docs server (run in your terminal)
claude mcp add --transport http claude-code-docs https://code.claude.com/docs/mcp

# or add a local browser tool as a stdio subprocess
claude mcp add playwright -- npx -y @playwright/mcp@latest

# Step 2: check the connection status
claude mcp list
Add a hosted server or a local tool, then list your servers to check the status.

Here is what claude mcp list shows you. Each server appears with its address and a status. Connected means it is ready. If you see Needs authentication, run /mcp inside a session to sign in.

$ claude mcp list
Checking MCP server health...

claude-code-docs: https://code.claude.com/docs/mcp (HTTP) - ✔ Connected
playwright: npx -y @playwright/mcp@latest (stdio) - ✔ Connected
sentry: https://mcp.sentry.dev/mcp (HTTP) - ! Needs authentication
The status at the end of each line is what you are checking: Connected is ready; Needs authentication wants a /mcp sign-in.

That is the whole loop: add, list, use. Once claude mcp list shows the server as Connected, Claude can work from that tool directly in every session. Add one server you copy data from every day, and see the difference.

Next post: an MCP deep dive — transports, config as code, and connecting your real systems.

MCP — How Claude Code Connects to Your Real Tools · Ali Huzaifa