How to Add MCP to Claude Code and Desktop

Understanding MCP and Claude Integration

MCP means Model Context Protocol. It gives Claude a shared way to use outside tools and data.

An MCP server offers tools, files, or records to an MCP client. Claude can then search data or take approved actions.

The MCP architecture guide explains how clients, servers, and tools fit together.

The short answer to “how to add MCP to Claude” is this. Add a server through a command or a JSON settings file.

Claude starts the server when it needs one. The server sends back results in a shared format.

Why Add MCP Servers?

MCP can give Claude fresh data instead of old pasted notes. A code server can search a repository or inspect an issue.

A work server can find tickets, pages, or team records. Claude can also link steps across tools.

These benefits make MCP useful for both small tasks and daily work:

• Search current files, code, and work records
• Reduce copy and paste work
• Run repeat tasks with clear approval points
• Keep tool access in one shared setup
• Limit access through narrow server permissions

Start with one low-risk server. Test it before adding more tools.

Server quality still matters. A poor server may return stale data or request broad access.

How to Add MCP Servers to Claude Code

If you need to know how to add MCP to Claude Code, use the command line first. It is quick, clear, and easy to check.

First, choose a trusted server. Find its launch command or its remote endpoint.

1. Pick a short server name, such as github-read.
2. Choose local stdio or remote HTTP transport.
3. Add the command or endpoint to Claude Code.
4. Set any needed values through environment variables.
5. List the server and test one small request.

For a local server, run this example in your shell:

claude mcp add --transport stdio my-tools -- npx -y @example/mcp-server

Replace the name and package with the values from your server guide. Keep the name unique.

For a remote server, use an HTTP endpoint:

claude mcp add --transport http team-data https://example.com/mcp

Run claude mcp list after saving the entry. This is the core answer for how to add MCP servers to Claude Code.

It also covers “claude code how to add MCP” and “how do I add MCP servers to Claude Code?”

Provider Examples for Claude Code

Provider setup follows the same pattern. The server guide supplies the package, endpoint, and access scope.

For GitHub, users often search how to add GitHub MCP to Claude Code. Pick a server that supports your needed GitHub actions.

For Atlassian, use the same check when asking how to add Atlassian MCP to Claude Code. Confirm whether it covers Jira, Confluence, or both.

Other common searches include how to add Jira MCP to Claude Code and how to add Linear MCP to Claude Code. The steps stay the same.

• Check the server owner and source code
• Review each tool before granting access
• Use read-only access for the first test
• Limit access to one repo, team, or project
• Save write access for tasks that need it

Not every provider uses the same login flow. Some use a token, while others use a browser sign-in.

Never place a live token in a public file. Store it in your shell or a trusted secret store.

JSON Settings, Names, and Authentication

JSON setup helps when you need more control. The file path can vary by tool version and setup scope.

Check the current Claude Code help output before editing a file. A typical local entry looks like this:

{"mcpServers":{"github-read":{"command":"npx","args":["-y","@example/github-mcp"],"env":{"GITHUB_TOKEN":"${GITHUB_TOKEN}"}}}}

The command starts the server. The args list passes launch options.

The env block passes values at run time. It should not hold the real secret.

A remote entry may look like this:

{"mcpServers":{"team-data":{"url":"https://example.com/mcp","headers":{"Authorization":"Bearer ${MCP_TOKEN}"}}}}

Use names that show both the tool and its risk. Names like jira-read beat vague names like server-one.

Save the file, then start a new Claude Code session. Run claude mcp list again.

How to Add MCP to Claude Desktop and Cursor

Claude Desktop uses a separate settings file from Claude Code. Its file still uses an mcpServers object.

That answers how to add MCP to Claude Desktop and how to add MCP servers to Claude Desktop. Find the current file path in Anthropic’s setup guide.

Paste the server entry, save the file, and fully restart Claude Desktop. A new chat may then show the server tools.

Cursor also has its own MCP settings. People often ask how to add MCP servers to Cursor.

For GitHub, that becomes how to add GitHub MCP to Cursor. For Jira, it becomes how to add Jira MCP to Cursor.

• Open the MCP settings in the app
• Add the server command or remote URL
• Set secrets outside the shared file
• Save the settings and restart the app
• Check that the expected tool appears

Do not copy a Claude Code file into Cursor without checking its format. Each client may support different fields.

Testing and Troubleshooting MCP Connections

Test one server before you add another. Ask for a simple read action with a narrow scope.

Then confirm the result against the source system. This catches stale data and wrong account access.

Use this quick fault check:

Problem	Likely cause	First check
Server is missing	Bad file or scope	Run the list command
Launch fails	Wrong command or package	Run it in the shell
Login fails	Missing or expired token	Check the environment value
Tools do not show	Old app session	Restart the client
Request is denied	Scope lacks access	Review account permissions

Read the exact error text before changing settings. Small spelling errors can stop a server from loading.

Check quotes, commas, braces, and field names in JSON. Use a trusted JSON checker if the file will not load.

For deeper client details, use Anthropic’s Claude Code MCP guide. It is the best source for current commands.

Manage Several Servers Safely

Several servers can work well when each has one clear job. Keep names short and group them by use.

Read access should be the default. Add write access only after a small, safe test.

Server name	Good use	Safer first scope
github-read	Search code and issues	One repository
jira-read	Find project work	One project
linear-read	Review team tasks	One team
local-files	Read work files	One folder

Keep secrets out of source control and shared settings. Rotate tokens when a team member leaves.

Review server tools from time to time. Remove entries that you no longer use.

Conclusion and Next Steps

The answer to how to use MCP servers with Claude Code starts with one small server. Add it, list it, and test one safe request.

The same approach works for Claude Desktop and Cursor. Their settings paths and supported fields may differ.

Common provider searches include how to add Notion MCP to Claude Code, how to add Puppeteer MCP to Claude Code, and how to add Slack MCP to Claude Code.

Use the provider’s current guide for each package. Confirm its access rules before you connect it.

Good setup is simple. Clear names, narrow access, safe secrets, and fresh sessions prevent most trouble.