SkyFi SkyFi

SkyFi MCP — Service Documentation

The SkyFi MCP server exposes SkyFi's satellite-imagery workflows to AI agents and other Model Context Protocol clients. Connect an MCP-capable agent, sign in with your SkyFi account, and the agent can search existing images (past captures in SkyFi's catalog), build areas of interest, plan and price tasking (new captures), place orders, pull deliverables, and run analytics — all on your behalf.

SkyFi offers imagery two ways: existing images — past captures already held in SkyFi's searchable catalog (the "archive"), bought with skyfi_place_archive_order — and tasking, which commissions a brand-new capture of an AOI in a future window. Existing images are immediate and lower-cost; tasking is for when nothing suitable already exists.

Heads up: some operations place real, billable orders and charge your SkyFi payment method or credits. See Billing before you let an agent order anything.

Connecting

MCP endpoint

Transport Streamable HTTP
Endpoint https://mcp.skyfi.com/mcp
Auth OAuth 2.1 bearer token (see below)

Point your MCP client at https://mcp.skyfi.com/mcp. The server advertises its OAuth configuration through the standard discovery documents, so most clients will walk you through sign-in automatically.

Authentication

The server implements a narrow slice of OAuth 2.1:

Discovery documents:

When a client starts the flow you are redirected to a SkyFi sign-in page, where you authenticate with your SkyFi email and password and accept the Terms of Service. The server then mints an authorization code and returns it to your client.

Scopes

Scope Grants
skyfi:read Search existing images, geocode and measure AOIs, read orders and deliverables
skyfi:write Manage saved AOIs and other account-scoped resources
skyfi:order Place orders and run paid analytics

Setting up a client? The setup tutorial walks through connecting from Claude Desktop, Claude Code, and Claude Team/Enterprise organizations, with screenshots and video walkthroughs.

Capabilities

Tools are grouped by what they do. Names are the tool ids your agent calls.

Account

Existing images are past captures SkyFi already holds in its catalog (the archive). Search them, pick one, then buy it with skyfi_place_archive_order. You buy only the overlap between your AOI and the image's footprint — that intersection is what gets cropped, delivered, and billed. Anything in your AOI outside the footprint isn't covered; use skyfi_aoi_coverage_gaps to find those gaps and task them.

Areas of interest (AOIs)

Tasking (new captures)

Validate & price (no charge)

Place orders (billable)

Orders & deliverables

Analytics (billable)

Saved AOIs

End-to-end automation

The groups above compose: an agent can chain them into a complete workflow from a single prompt. For example:

"Run NDWI on the latest VANTOR image over my Rotterdam port saved AOI."

The agent drives the whole pipeline automatically — no manual steps in between: it lists your saved AOIs (skyfi_list_saved_aois) and finds the one covering the port of Rotterdam, searches the catalog for the most recent VANTOR capture over that AOI (skyfi_search_archive), validates the price and purchases the image (skyfi_validate_archive_orderskyfi_place_archive_order), then orders an NDWI standing-water analysis on the delivered image (skyfi_ndwi_standing_water). The only thing left to you is approving the two charges — image purchase and analytic — when the agent presents the exact prices.

the end-to-end run: saved AOI → latest VANTOR image → purchase → NDWI result
the end-to-end run: saved AOI → latest VANTOR image → purchase → NDWI result
the end-to-end run: saved AOI → latest VANTOR image → purchase → NDWI result
Given we've the real order capability, ordering of the existing image and running NDWI on top would be automated.

Billing and the validate-before-charge flow

Several tools charge the customer's real payment method or credits and are not reversible: skyfi_place_archive_order, skyfi_place_tasking_order, skyfi_ndvi_vegetation_health, skyfi_ndwi_standing_water, skyfi_building_detection, skyfi_vessel_detection, and skyfi_vehicle_detection. skyfi_tasking_order_form also places a real order when the user clicks Execute.

Before any charge, the workflow is:

  1. Validate. For existing-image/tasking orders, call skyfi_validate_archive_order / skyfi_validate_tasking_order first — they return the authoritative price and confirm the order is placeable. The place tools re-validate and refuse to charge if validation fails. Analytics tools return a dry-run price when called with confirm=False.
  2. Confirm. Show the customer the exact total and get explicit, unambiguous confirmation that they want to be charged that amount.
  3. Charge. Only then call the tool with confirm=True.
  1. Call skyfi_whoami first to confirm authentication and remaining budget.
  2. For any place name, call skyfi_geocode_aoi to get a WKT polygon — every other tool expects WKT in EPSG:4326.
  3. Before a tasking order, run skyfi_tasking_cloud_forecast for the window and show the daily cloud breakdown; propose a lower-cloud window if one exists.
  4. Validate, confirm, then place.

Support

Questions or issues? Contact [email protected] or visit skyfi.com.

SkyFi · skyfi.com · Setup tutorial · Documentation · Terms of Service · Privacy Policy