MongoDB MCP Server

What it does

The MongoDB MCP Server connects AI coding agents to MongoDB deployments (Atlas, Community Edition, or Enterprise). Agents can explore collections, run queries, and generate code that matches your actual schema and data. Supports natural-language questions about your data and produces optimized queries and implementation code.

Available tools

Tool	What it does
list_databases	List databases in the deployment
list_collections	List collections in a database
run_query	Execute a read query and return results
get_schema	Infer schema from collection documents
explain_query	Explain query plan for a given query

Exact tool names may vary; check the MongoDB MCP docs for the current API. The server can run in read-only mode with --readOnly.

Setup by tool

Cursor

Create .cursor/mcp.json:

{
  "mcpServers": {
    "mongodb": {
      "command": "npx",
      "args": ["-y", "mongodb-mcp-server@latest", "--readOnly"],
      "env": {
        "MDB_MCP_CONNECTION_STRING": "mongodb+srv://user:pass@cluster.mongodb.net/mydb"
      }
    }
  }
}

For Atlas API credentials instead of a connection string:

{
  "env": {
    "MDB_MCP_API_CLIENT_ID": "your-client-id",
    "MDB_MCP_API_CLIENT_SECRET": "your-client-secret"
  }
}

Claude Code

Add the same configuration to your Claude MCP config. Replace the connection string with your deployment URL. Use --readOnly when agents only need read access.

VS Code / GitHub Copilot

Add the server with the same command, args, and env. Ensure Node.js v20+ is available.

When to use this

• Query generation: Agents write MongoDB queries against real collections and schema
• Schema discovery: Infer schema from documents before generating models or ORM code
• Data exploration: Ask natural-language questions about your data during development
• API development: Generate CRUD endpoints that match your document structure
• Migration planning: Inspect existing collections before schema changes

Security considerations

• Use --readOnly when agents only need to read data
• Connect to development or staging clusters, not production
• Connection strings contain credentials; use environment variables, not committed files
• Atlas API credentials should have minimal required scopes
• Consider network restrictions so only authorized hosts can reach the deployment
• Audit queries run by agents; avoid exposing sensitive collections unnecessarily