mcp.json
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"."
]
},
"github": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-github"
]
}
}
}
Use the MCP Server Config Generator to create a small starter mcp.json for AI coding tools. Start with the servers that remove repeated context gathering, then expand only after checking permissions and credentials.
How To Choose MCP Servers
Choose servers by workflow need. Filesystem and GitHub are good starting points, browser automation helps frontend checks, and database servers should usually start read-only.
Before Sharing mcp.json
Review every command, token boundary, and permission scope. Keep secrets in environment variables and document who owns each server.
FAQ
Should I add every available MCP server?
No. A small reviewed setup is easier to maintain than a large fragile config.
Can one mcp.json work everywhere?
The intent can be shared, but exact paths and config locations can differ by tool.