Claude Code MCP: How to Connect an MCP Server to Claude Code

What is Claude Code?

Claude Code is a command-line agent built by Anthropic. You run it in your terminal, point it at a project, and ask it to do work in plain language. It reads and writes files, runs commands, and chains several steps to complete a task, with you approving along the way.

Out of the box, it works on the code and files in your working directory. The Model Context Protocol is what extends it beyond that, to the systems where the rest of your work lives.

What is an MCP server, and why connect one?

An MCP server is a small bridge that exposes a tool’s capabilities, like a CRM’s records or an outreach platform’s campaigns, through the open Model Context Protocol. Any MCP-aware client, Claude Code included, can connect to it and use those capabilities.

Connecting one to Claude Code matters because it closes the gap between writing about your work and doing it. Instead of describing the data you want and copying it back and forth, you let Claude Code reach the source, read it, and act on it in the same conversation. It works the same way across Claude, ChatGPT (OpenAI), and any other assistant that supports MCP.

How to connect an MCP server to Claude Code

Adding a server takes one command. Claude Code supports three transports, depending on how the server runs.

• Add the server. Use the claude mcp add command. For a remote server over HTTP, run claude mcp add --transport http . For a server that streams over SSE, use --transport sse. For a local server that runs as a process on your machine, pass the command after --, like claude mcp add -- .
• Pick a scope. A server is local to you by default. Use -s project to share it with your team through a .mcp.json file committed to the repo, or -s user to make it available across all your projects.
• Authenticate. Inside a Claude Code session, run the /mcp command to see your servers and their status, and to complete OAuth for the ones that need it.
• Use it. Once connected, just ask. Claude Code sees the server’s tools and calls them as part of its work.

MCP is host-agnostic, so the same server works in other assistants too, which we cover just below.

Add an MCP server in ChatGPT and Codex

Because MCP is an open standard, the same server you added to Claude Code works in OpenAI’s tools too. There are two common paths.

• ChatGPT, with connectors. In ChatGPT, open Settings, go to Connectors, and add a custom connector by pasting the server’s URL, then authenticate. ChatGPT exposes the server’s tools inside your chats. See OpenAI’s connectors guide.
• OpenAI Codex. Codex, OpenAI’s coding agent, reads MCP servers from its config with the codex mcp add command, much like Claude Code. Add the server there and Codex can call its tools as part of a task. See the Codex MCP docs.

One thing to watch: a remote HTTP or SSE server connects almost anywhere, while a local stdio server needs a host that can launch a process on your machine. Pick the transport the host supports, and you can run the same La Growth Machine or CRM server across Claude Code, ChatGPT, and Codex.

Why use an MCP with Claude Code

The point of an MCP is to put Claude Code where your real work happens. For a go-to-market team, that means connecting the tools that hold your pipeline and run your outreach, then driving them from one prompt.

With the La Growth Machine MCP connected, you can do this from the terminal:

• Build an audience from a LinkedIn or Sales Navigator search.
• Launch a multichannel sequence across LinkedIn and email.
• Read and triage replies without opening the app.

Chain that with a CRM MCP and an enrichment MCP, and a single instruction can go from “find these people” to “they are enriched, in the CRM, and in a sequence.” Claude Code stops being a coding assistant and becomes the place you pilot the stack, moving from executor to operator to strategist.

Tips and security

An MCP server can reach real systems, so connect it deliberately.

• Mind the scope. Use project scope only for servers your whole team should share, and review a .mcp.json before you commit it.
• Confirm before writes. Let Claude Code read and search freely, but confirm before it creates, updates, or sends anything.
• Use trusted connectors. Prefer official servers, and treat community ones with caution, since a server runs with the access you give it.
• Connect a least-privilege account. The agent inherits whatever the connected account can do.

Frequently asked questions