Skan MCP server - Skan

Skan exposes its skin analysis and product catalog to AI assistants as a remote MCP server. A client starts a scan from a person’s photo, polls for the result, and then matches products and explains ingredients against that scan.

Endpoints

Endpoint	URL	Auth
Guest	`https://skan-connectors-staging.skan.workers.dev/mcp`	None
Account	`https://skan-connectors-staging.skan.workers.dev/mcp/account`	OAuth 2.1 (details)

These are the staging endpoints. The production endpoint will replace them here when it is live.

Transport: Streamable HTTP. JSON-RPC 2.0 over POST, with responses as JSON or an SSE stream.
Protocol version: 2025-06-18.
Session: stateless. No Mcp-Session-Id is issued, and any client can use a scan by passing its run_id and access_token.
Capabilities: tools only (listChanged: false). There are no resources or prompts.

Tools

Tool	What it does	Completes
`start_skin_scan`	Starts a scan, from an inline photo or through an upload link	Immediately; analysis runs in the background
`get_skin_scan`	Polls a scan and returns scores when complete	Immediately
`delete_skin_scan`	Deletes a guest scan and revokes its token	Immediately
`recommend_products`	Scores catalog products against a completed scan	Immediately
`find_product`	Looks up a named product or brand, scored against the scan	Immediately
`analyze_ingredients`	Explains an ingredient list	Immediately
`delete_my_skan_data`	Deletes everything held for a signed-in account	Immediately; account endpoint only

Typical order: start_skin_scan, then get_skin_scan until completed, then any of the catalog tools. See scan lifecycle.

Results are cosmetic appearance measurements, not a diagnosis. See cosmetic limits.