使用方法
- 复制安装命令，在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物，并对照来源检查输出。
npx skills add https://github.com/cartodb/agent-skills --skill carto-explore-datawarehouse开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况，不等于独立质量或安全测试。
仓库原始内容
按源仓库内容呈现，保留标题、案例、代码、表格、链接以及原文引用的演示图片。
carto-explore-datawarehouse
Before writing SQL or building maps, an agent usually needs to know what's in the warehouse. This skill covers discovery, matched across MCP and CLI:
| Task | MCP (primary) | CLI (fallback) |
|---|---|---|
| List / inspect connections | explore_data (list_connections, get_connection) | carto connections list / get |
| Walk the hierarchy (project → dataset → table) | explore_data (list_resources) | carto connections browse |
| Full-text find a resource | explore_data (search) | — |
| Columns + types for a table | explore_data (describe) | carto connections describe |
| Find / inspect a named source | manage_named_sources (list, get) | carto named-sources list / get |
Access-path routing.explore_dataworks on any attached MCP session, but not all its methods do on a token:list_connections,list_resources, andsearchneed only the MCP Server scope, while `describe` additionally needs the Maps API on a token session (it's always available over OAuth).manage_named_sourcesis an authoring tool, so it needs an OAuth session (it's hidden on token sessions). Use thecarto connections/carto named-sourcesCLI when the server isn't attached, when a token lacks the scope fordescribe, or when the exploration is scripted/headless. Detection signals and the full routing table: `carto-basics/references/access-paths.md`.
When to use this skill
- You don't know which tables / schemas exist in a connection.
- You need a column list and types before writing SQL or authoring a map.
- The user references "the named source for X" and you need to find it.
If you already know the table and just want to query it, jump to `carto-query-datawarehouse`.
Path syntax by engine
The path passed to list_resources / describe (or connections browse / describe) depends on the engine:
| Engine | Path shape |
|---|---|
| BigQuery | project.dataset.table |
| Snowflake | DATABASE.SCHEMA.TABLE |
| Postgres / Redshift | schema.table (no leading project/database) |
| Databricks | catalog.schema.table |
CLI examples:
bash
carto connections browse <name> # top level
carto connections browse <name> "carto-demo-data.demo_tables" # drill in
carto connections describe <name> "carto-demo-data.demo_tables.nyc_collisions"What's in this skill
| Topic | Reference |
|---|---|
| Browsing and describing a connection in detail | references/connection-browse.md |
| Named sources — what they are, how to list and inspect them | references/named-sources.md |
Always-on guidance
- Browse before you query. A two-second
list_resourcesusually saves a five-minute "table not found" loop. - `describe` returns column types — use them to write correct SQL (e.g. don't
ST_DWithinagainst aSTRINGcolumn mistakenly namedgeom). - Paginate long lists — CLI
browsedefaults to--page-size 30; bump it for datasets with hundreds of tables. - Named sources ≠ tables. They're parameterized queries — inspect the underlying tables before assuming a column you see in the source exists in raw form.
- `carto-demo-data` is a public BigQuery dataset CARTO ships — browsing it works on any BigQuery connection with the right IAM, a fast way to validate a fresh connection without touching customer data.
