Rclone

You can run a Model Context Protocol (MCP) server for the Rclone RC API to control cloud storage remotes, copy and sync files, list directories, and perform other operations through natural language. This enables AI assistants to manage your rclone setup securely and efficiently by exposing a curated set of endpoints that your MCP client can invoke.

How to use

To use the MCP server, start a local MCP instance that talks to your running rclone RC daemon. Create an MCP client configuration that points to the MCP server you run, then issue natural language requests like “list files in remote X” or “copy from remote A to remote B.” You can enable authentication on the MCP server to restrict access and pick which toolsets are exposed. The server exposes endpoints via a toolset arrangement, so you can tailor access to only the capabilities you need.

How to install

Prerequisites you need before installing include a running rclone RC daemon and a Node.js environment for running the MCP server binary.

1) Ensure an rclone RC daemon is running locally

rclone rcd --rc-no-auth

# or with authentication:
rclone rcd --rc-user=admin --rc-pass=secret

2) Install and run the MCP server using npx

# Default setup (no auth)
npx rclone-mcp

# With authentication (provide admin credentials in the environment)
RCLONE_URL=http://localhost:5572 RCLONE_USER=admin RCLONE_PASS=secret npx rclone-mcp

3) Optional: expose via standard http transport for remote clients

npx rclone-mcp http --port 3000

4) If you are using Docker, you can build and run the container

docker build -t rclone-mcp .

docker run -i --rm -e RCLONE_URL=http://host.docker.internal:5572 rclone-mcp

Configuration and usage notes

The MCP server accepts configuration through environment variables and command-line options. Use the following environment variables to control access and tool exposure.

Environment Variables

| Variable | Description | Default |
|---|---|---|
| RCLONE_URL | rclone RC daemon URL | http://localhost:5572 |
| RCLONE_USER | HTTP Basic Auth username | — |
| RCLONE_PASS | HTTP Basic Auth password | — |
| RCLONE_TOOLSETS | Comma-separated toolset list | default |
| RCLONE_READ_ONLY | Set to 1 to disable write tools | — |

Operational tips

• Use the default toolsets to expose core configuration and operations capabilities. You can also selectively enable or disable toolsets to keep the available actions focused for AI assistants.
• Enable read-only mode to restrict the MCP server from performing mutating actions such as copy, move, delete, or sync. This is useful for environments where AI is allowed only to view data.

Security considerations

• If you enable authentication, provide a strong username and password and keep them secret. Treat RCLONE_URL and credentials as sensitive data.
• Consider exposing the MCP server only inside trusted networks or behind a reverse proxy with proper TLS and access control.

Troubleshooting

• If the MCP server cannot communicate with the rclone RC daemon, verify that the daemon URL in RCLONE_URL matches where rclone rcd is listening.
• Ensure the MCP process has network access if running in containers or remote environments.

Examples

Basic local setup (no auth):

• Start rclone RC daemon if needed
• Run MCP server with default toolsets
• Optional: use http transport for remote clients Example commands are shown in the installation section above.