How to add an MCP server to Claude Code
Adding an MCP server to Claude Code takes one command, but the flags matter: local versus remote transport, and which scope the config lands in. Here is the whole process, plus a worked example install you can copy verbatim.
What an MCP server adds
The Model Context Protocol (MCP) is an open standard that lets Claude Code talk to outside tools and data: databases, APIs, browsers, project trackers, or research pipelines. An MCP server exposes a set of tools; once connected, Claude can call them mid-task like any built-in capability. Connecting one takes a single command.
The one command: claude mcp add
Claude Code manages servers with the claude mcp CLI:
- Local (stdio) server:
claude mcp add my-server -- npx -y some-mcp-package; everything after--is the command that starts the server. - Remote (HTTP) server:
claude mcp add --transport http my-server https://example.com/mcp. - Housekeeping:
claude mcp list,claude mcp get my-server, andclaude mcp remove my-server.
Pick the right scope
The --scope flag controls who sees the server:
- local (the default): just you, just this project.
- project: written to a
.mcp.jsonfile at the repo root, checked in, and shared with your team. - user: available to you across every project; right for personal tools you use everywhere.
Remote servers and authentication
Remote servers are the lowest-friction option: nothing to install or keep running, just a URL. Some authenticate with OAuth, which you complete by running /mcp inside Claude Code and following the login flow. Others use an API key passed as a header: claude mcp add --transport http my-server https://example.com/mcp --header "X-Api-Key: your-key".
Worked example: the give-back MCP
Here is a real install, end to end, using Tokens for Good, the MCP server that turns spare Claude capacity into nonprofit research. The easiest path is the npm initializer, which connects the server and installs its skills in one step:
npx tokens-for-good init
Or add the remote endpoint directly at https://tokensforgood.ai/mcp with your API key header; the docs walk through both. Once connected, /tfg researches one nonprofit and /tfg-schedule makes it recurring. Every report feeds a public directory, and it all runs on the subscription you already pay for.
Verify it works, and keep the list short
Run /mcp in a session to see connected servers, their status, and their tools; if a server shows as failed, claude mcp get prints the config to sanity-check. One practical warning: every active server adds tool definitions to your context, so clients get unwieldy past a handful. Enable the two or three that match what you are doing; some curated picks are in useful MCP servers and MCP servers for good.
Frequently asked questions
How do I add a remote MCP server to Claude Code?
Where does Claude Code store MCP configuration?
What is the difference between a local and a remote MCP server?
How many MCP servers should I add?
Add the MCP server that gives back
Tokens for Good connects in about a minute and turns spare Claude capacity into verified nonprofit research.
Install Tokens for Good