Source Catalog Integration (V1.5)

Forge data products directly from your existing catalog metadata — no re-typing descriptions, tags, lineage, ownership, or sensitivity classifications you've already invested in. V1.5 turns forge-cli into a vendor-neutral modeling layer that consumes the seven major catalog ecosystems and emits one Fluid contract.

fluid forge data-model from-source \
  --source snowflake \
  --credential-id snowflake-prod \
  --database BIZ_LAB --schema SEEDED \
  --technique data-vault-2 \
  -o biz_lab.fluid.yaml

This page is the catalog-by-catalog index. Each linked page is a self-contained walkthrough: privileges to grant, env-vars to set, auth methods, end-to-end demo, and error reference.

Supported catalogs

Combined: pip install "data-product-forge[catalogs]" installs every optional dep at once. Default pip install data-product-forge ships with zero catalog deps — you opt into only the catalogs you use.

Two ways to drive a catalog forge

1. CLI: fluid forge data-model from-source

fluid forge data-model from-source \
  --source <catalog>               # snowflake, unity, bigquery, dataplex, glue, datahub, datamesh_manager
  --credential-id <name>           # saved name from ~/.fluid/sources.yaml
  --database <db> --schema <sch> \
  --tables orders customers \
  --technique data-vault-2 | dimensional \
  -o contract.fluid.yaml

Non-interactive; reproducible; works in CI. The --source value selects the catalog adapter; --credential-id points at a row in ~/.fluid/sources.yaml configured via fluid ai setup --source CATALOG --name NAME.

2. MCP: forge_from_source

// .mcp.json (Claude Code, Cursor, any MCP client)
{
  "fluid": {
    "command": "fluid",
    "args": ["mcp", "serve"],
    "env": { "SNOWFLAKE_PRIVATE_KEY_PATH": "/path/to/key.p8" }
  }
}

Then tell the agent:

Forge a data-vault-2 model from my Snowflake BIZ_LAB.SEEDED schema. Use the snowflake adapter with credential id snowflake-prod. Save the contract to biz_lab.fluid.yaml.

The agent calls the MCP forge_from_source tool; same pipeline, same output.

Setup wizard: fluid ai setup --source CATALOG --name NAME

For first-time setup of a new source:

fluid ai setup --source snowflake --name snowflake-prod
# ? Account: myorg-abc12345
# ? Auth method:                 # ★ marks recommended path
#   ★ key_pair (recommended)     # — most secure, no password
#     password
#     externalbrowser (SSO)
# ? Private key path: /Users/me/.ssh/snowflake_key.p8
# ? Default warehouse: COMPUTE_WH
# ? Default role: ANALYST
# ✓ Saved to ~/.fluid/sources.yaml

This writes a single row to ~/.fluid/sources.yaml. Secrets land in the OS keyring (macOS Keychain / Windows Credential Manager / Linux secret-service); only non-sensitive fields land on disk.

Three guarantees that hold across every catalog

1. Read-only. Adapters request INFORMATION_SCHEMA-equivalent privileges only. No SELECT * against any data table — only metadata.
2. No data values fetched. Tag values, descriptions, classifications, schema — never row data. Defends against an upstream agent escalating "list tables" into "dump customer PII."
3. Audited. Every catalog tool call writes an audit event via ~/.fluid/store/audit/. Credentials are scrubbed before write. Use fluid memory show audit to query.

Auto-detection: fluid ai status lists what you have

fluid ai status
# Configured sources:
#   snowflake-prod  (snowflake / key_pair, ANALYST role)
#   unity-staging   (databricks / pat)
#   datahub-corp    (datahub / pat)

When you run fluid forge without --source, AI mode's interview auto-suggests the configured source — same UX as fluid ai setup auto-discovering provider env vars.

Error model

Every catalog adapter raises one of three typed exceptions, each carrying a suggestions: list[str] field with the next-action:

The next-action is in the message — not buried in the docs.

See also

• Forge data-model overview
• MCP server walkthrough
• Add your own catalog adapter