fusebase-dashboards

Call getDashboardView first to get schema.items with column keys, then use getDashboardViewData to read rows or batchPutDashboardData to create and update rows. Always use the opaque item_key from the schema, never the column display name.

Start with getDashboards({ scope_type, scope_id, compact: true }) with no other filter to list every accessible dashboard. Name and alias filters are exact matches, so fall back to the unfiltered list or resolveAliases when a lookup returns empty.

Use MCP tools (tools_search, tools_describe, tool_call) for actions inside the LLM session during development. Use SDK methods only when generating runtime code for the feature or browser; never mix both chains in one scenario.

An empty result usually means the connection token is dashboard-scoped, granting dashboards but not databases. This is an access boundary, not an empty organization; use the plain getDashboards list instead.

Create rows with batchPutDashboardData while omitting lookup columns, then call findRelationsByDashboardIds to get the relationId and addRelationRows with source_index and target_index row IDs. Writing lookup columns directly via batchPutDashboardData causes a 400 error.

It does not cover file upload lifecycles, Gate file operations, or runtime SDK implementation details, which belong to the file-upload, fusebase-gate, and SDK skills. It is also not the default storage-planning skill for new app-owned structured data.