Marmot: a data catalog skill

---
name: marmot
description: Interact with a Marmot data catalog instance. Use this skill when the user wants to search for data assets, view lineage, browse glossary terms, check pipeline runs, manage tags or owners, view metrics, query catalogued data, or do anything related to their data catalog. Covers the Marmot CLI, REST API and MCP server.
---

This skill helps you interact with a Marmot data catalog. Marmot catalogs data assets (databases, tables, topics, APIs, dashboards) across an organisation's data stack and tracks lineage between them.

## Setup

The user needs the Marmot CLI installed and authenticated. If not already set up:

```bash
curl -fsSL get.marmotdata.io | sh
marmot login https://marmot.example.com
```

`marmot login` opens a browser for OAuth 2.0 PKCE authentication and caches the token locally. The token expires after 24 hours and carries the user's roles and permissions as claims. Login also registers the token with the Docker credential store for the instance's host, so `oras push`, `crane push` and `docker push` can publish plugins to the registry the instance serves. This goes through the `docker-credential-marmot` helper, a symlink to the marmot binary that `install.sh` creates; if the helper is not on your PATH, login writes the token itself into `~/.docker/config.json` the way `docker login` does. Alternatively, use an API key:

```bash
export MARMOT_HOST=https://marmot.example.com
export MARMOT_API_KEY=<key>
```

Or pass `--host` and `--api-key` as flags on any command.

`marmot skills install` refreshes this skill to match the installed CLI.

## MCP Server

Marmot has a built-in MCP (Model Context Protocol) server. If the user has configured MCP, you can interact with the catalog directly using these tools instead of the CLI:

Reading the catalog:

- **discover_data**: unified data discovery. Supports natural language queries, lookups by ID or MRN (e.g. `postgres://db/schema/table`), filtering by type, provider and tags, and metadata-based queries.
- **find_ownership**: ownership in both directions. "Who owns this asset?", "What does this user own?", "Show all data owned by the data-eng team". Works for assets and glossary terms.
- **lookup_term**: business glossary lookups. Search for terms by name or retrieve definitions.
- **explore_data_products**: curated collections of assets that serve a business purpose, and what they contain.
- **explore_teams**: teams and their members.
- **trace_lineage**: what feeds an asset (upstream) and what it feeds (downstream).

Changing the catalog (only offered to callers allowed to make the change):

- **update_documentation**: write or clear the human-authored description on an asset.
- **manage_tags**: add or remove tags on an asset.
- **manage_owners**: add or remove a user or team as an owner of an asset.

Querying data (only offered to callers with `assets:query`):

- **list_queryable_sources**: the sources that answer queries, searchable by name.
- **query_data**: run one read-only SQL statement against a source. The answer carries the rows plus what the catalog knows about every table read, and the id the query was recorded under.

Ask before calling the tools that change the catalog, even when the user's request implies the change. Say what you are about to write and to which asset.