MCP Server

The CLI doubles as a local MCP server, exposing your media library to AI agents (Claude Desktop, Cursor, and other MCP clients) over stdio — using the API key in your config.

Run the server

medialocker mcp serve

This speaks the MCP protocol on stdio, so it's launched by an MCP client rather than run by hand. stdout is reserved for the protocol; logs go to stderr.

Install into a client

medialocker mcp install --client claude          # print a config block
medialocker mcp install --client claude --write  # merge into Claude Desktop's config
medialocker mcp install --client cursor          # print a Cursor config block

install emits (or merges) a medialocker MCP server entry that runs medialocker mcp serve with your resolved URL and API key.

Tools

The local server exposes these tools, each scoped by your API key's permissions:

Tool	Does
list_buckets	List buckets
create_bucket	Create a bucket
list_media	List/filter media objects
search_media	Full-text + facet search
get_object	Object metadata by id
get_object_url	A presigned download URL
upload_object	Upload a local file to a bucket
get_usage	Storage and request usage snapshot

Local CLI server vs. the hosted MCP

MediaLocker also runs a hosted MCP server at https://mcp.medialocker.io over Streamable HTTP, with a richer tool catalog and a multi-tenant gateway. Pick based on how the agent connects:

	Local (medialocker mcp serve)	Hosted (mcp.medialocker.io)
Transport	stdio (runs on your machine)	Streamable HTTP (remote)
Setup	mcp install	Point a client at the URL with a bearer token
Tools	The core set above	The full catalog
Best for	Local/desktop agents, self-hosted instances	Always-on, multi-client access

TIP

Use the local server when an agent runs on the same machine as the CLI; use the hosted MCP server for remote or always-on access. Both honor the same API keys and scopes.

See also

• MCP Server — the hosted Streamable-HTTP server
• Command Reference — the rest of the CLI