Rowset Features

Rowset Features

Use this skill to explain Rowset's current feature surface accurately. Always prefer live discovery over memory:

Do not load capabilities or list datasets merely because a session started. Use the live discovery steps below when the user asks about Rowset's feature surface or when troubleshooting requires current capability details.

1. Load the compact available_topics index through a bare get_rowset_capabilities call, rowset capabilities command, or REST capabilities request.
2. Request only relevant topics for detailed guidance. Use cases are opt-in; full mode retrieves the complete guide.
3. Inspect current MCP schemas, CLI help, or generated REST API docs for the interface the user selected.
4. Use llms.txt from the Rowset site to find current public reference material.

The sections below are orientation, not a complete or permanent feature list. When they disagree with a live Rowset resource, follow the live resource.

Do not claim Rowset-owned Google Sheets sync, dashboard upload wizards, or spreadsheet write-back as active product capabilities.

Core Capabilities

MCP setup and verification

• Configure Rowset as a remote Streamable HTTP MCP server.
• Store the full API key in a private ROWSET_API_KEY environment variable.
• Verify auth with get_user_info only for a new or failing connection. If Rowset is already configured and authenticated, skip connection verification and the activation handoff.
• Load the topic index with get_rowset_capabilities, then request specific topics. Set include_use_cases only when examples help, or full for the complete guide.

Datasets

• When the relevant dataset is unknown, use search_datasets with a limit of 3; avoid listing unrelated datasets.
• Create ready API-backed datasets with create_dataset.
• Inspect one dataset with get_dataset before row work.
• Use rowset_id generation when no reliable business key exists.

Dataset context and semantic schema

• Persist description, instructions, and JSON metadata with update_dataset_metadata.
• Persist semantic column metadata with update_dataset_column_types.
• Supported column types: text, tags, choice, integer, number, currency, boolean, date, datetime, email, url, reference, and calculated.
• Use column descriptions when an agent should not infer meaning from a header.
• Use tags for comma-separated string values that should render as individual pills without changing row data returned through MCP or REST.
• Formula columns are read-only live values. Set calculation to formula, provide a result_type and formula expression, and use IF, SWITCH, AND, OR, NOT, DATEADD, TODAY, or NOW.

Schema changes

• Add columns with add_column.
• Rename columns with rename_column.
• Drop non-index columns with drop_column.
• Reorder columns with reorder_columns.
• Relationship columns, target index columns, and columns referenced by formulas must be unlinked or updated before destructive schema changes.

Relationships

Use dataset relationships when a source dataset column stores another dataset row's index value. For example, Messages.person_id can point to People.person_id.

• List relationships with list_dataset_relationships.
• Create relationships with create_dataset_relationship.
• Resolve a source row through a relationship with resolve_dataset_relationship.
• Delete relationship definitions with delete_dataset_relationship.
• With enforcement enabled, row writes fail when non-blank source values do not match an existing target row index.

Projects

• Discover projects with get_all_projects and search_projects.
• Create projects with create_project.
• Inspect a project and its datasets with get_project.
• Discover project sections with get_project_sections.
• Create sections with create_project_section.
• Update project name or description with update_project.
• Replace project JSON metadata with update_project_metadata.
• Update sections with update_project_section.
• Archive sections with archive_project_section.
• Move datasets with update_dataset_project, optionally passing section_key with project_key.

Projects organize datasets and carry workflow metadata. Sections optionally organize datasets inside one project. Neither changes authenticated access boundaries.

Rows

• List rows with search, filters, and sort through list_dataset_rows.
• Search rows across datasets with ranked hybrid retrieval through search_rows when vector search is enabled.
• Search rows within one known dataset through search_dataset_rows.
• Read rows with get_dataset_row or get_dataset_row_by_index.
• Create rows with create_dataset_row.
• Patch rows with update_dataset_row or update_dataset_row_by_index.
• Delete rows with delete_dataset_row only when the user requested deletion.

Public previews

• Use update_dataset_public_preview only when the user asks to share a read-only dataset.
• Enabled public datasets have a browser preview and dedicated read-only JSON metadata and row endpoints.
• Unprotected public datasets need no credential. Password-protected public API requests require X-Rowset-Public-Password on every request.
• Public access is not a replacement for authenticated MCP or REST when data is private or any write is required.

Archive, restore, and exports

• Archive datasets with archive_dataset without deleting rows or schema metadata.
• Restore archived datasets with restore_dataset.
• Use REST export endpoints for file snapshots: /export.csv, /export.jsonl, /export.xlsx, and /export.sqlite.

Answering Capability Questions

When explaining Rowset, keep the answer grounded in one of these categories:

• MCP setup and auth
• Dataset discovery and creation
• Dataset context and semantic schema
• Schema changes
• Relationships
• Projects
• Rows
• Public previews
• Archive, restore, and export

If a user asks for a capability outside these categories, say it is not an active Rowset product path unless the live MCP capability guide or API docs show otherwise.