How to call mcp servers in cursor?
Short answerYou do not call MCP servers directly in Cursor. Add the server in its MCP settings or an mcp.json file, turn it on, then ask the agent in chat to use its tools for your task.
Add the server config
Cursor reads MCP server definitions from a JSON file. You can add a project file inside the .cursor folder, or a global one in your user settings.
Each entry gives the server a name and either a command to run locally or a URL for a remote server.
- Open Cursor's MCP settings or the mcp.json file
- Add a name with a command or URL
- Put any API keys in the env section
- Save and check that the server shows as connected
Turn on the tools
Once the server is connected, its tools are listed in the settings panel, where you can enable or disable each one. Disabling tools you do not need keeps the agent focused.
If a server fails to start, its error message usually appears next to its name.
Use the tools from chat
Switch the chat to agent mode and describe the result you want, such as finding an issue or reading a file. The agent chooses the right tool and shows the call as it runs.
Review the output before you let the agent make changes based on that data.
Common mistakes
- Leaving a syntax error in the JSON, which stops the server from loading.
- Using a chat mode that does not run tools and then expecting results.
- Putting API keys directly in a config file you commit to git.
