Spade runs an MCP server, so an AI assistant can work in your Spade account. Describe the result you want in plain language, and the assistant finds the blocks, drafts the pipeline, checks it, and runs it once you say so.
You: Clip dem.tif to my study area and map its slope.
Assistant: I found gdal.clip_raster_by_vector, gdal.slope and
viz.raster_map, and read their inputs. The draft validates.
Shall I run it?Grounded in what exists
The assistant reads the real definition of every block before it uses one, and every pipeline goes through the same validator spade check uses before it can run. A misremembered block name or a miswired input is caught and fixed, not submitted.
Your account, your rules
You connect with your normal Spade sign-in, and the assistant acts as you: it sees your pipelines, runs and blocks, plus whatever is shared with you or public, and nothing more. It never sees a secret's value, it cannot publish collections, and every run it starts appears in the web UI alongside your own.
When something fails
The assistant follows each run block by block. If one fails, it reads that block's logs and tells you what went wrong, so you can fix the pipeline together and try again.
Connect
Add https://app.psae.us/mcp as a remote MCP server: a custom connector in Claude, or claude mcp add --transport http spade https://app.psae.us/mcp in Claude Code. The setup guide has the details.