dbt artifacts & target/

dbt-tools analyzes structured JSON outputs from dbt runs. The primary inputs are manifest.json and run_results.json at the root of a dbt target directory (local ./target or a remote prefix with the same file names).

Coding agents use agent skills that invoke CLI or MCP; skills do not read artifacts themselves.

Remote s3:// and gs:// prefixes use the same file names at the prefix root—see Local and remote artifacts.

When artifacts appear

dbt command	Typical artifacts	dbt-tools status
dbt compile	manifest.json (and related compile outputs)	manifest-only — graph and configs, no run results
dbt run, dbt build, dbt test	manifest.json + run_results.json	full — execution timing, status, and per-node results

If the Web UI is empty or execution views are missing, you may be on a manifest-only target. Run a command that produces run_results.json, or point at a directory that already has both files. See Troubleshooting and Web UI shows empty views.

What each file is (dbt-tools lens)

Artifact	dbt produces	dbt-tools uses it for
manifest.json	Project graph, configs, compiled SQL metadata	discover, explain, deps, lineage in Web
run_results.json	Per-node status, timing, adapter messages	Failed nodes, slow-run ranking, execution timeline
catalog.json	Warehouse column metadata	Optional enrichment when present
sources.json	Source freshness results	Optional; freshness checks are run by dbt, not dbt-tools

Schema versions depend on your dbt release. See Version compatibility for supported artifact versions—do not assume a single schema across all projects.

Target layout

Tools expect a dbt target directory (typically ./target) with manifest.json and run_results.json at the root of that location. Remote object storage follows the same one-pair-per-location contract—see Local and remote artifacts.

dbt Docs site vs JSON artifacts

Two different outputs are often confused:

• dbt docs generate + dbt docs serve — human-readable project documentation (HTML) for stakeholders and analysts.
• manifest.json / run_results.json — machine-readable artifacts under target/ that dbt-tools parses.

dbt-tools does not read the HTML docs site; it reads JSON artifacts from a local path or remote prefix. For building and viewing dbt Docs, see Build and view your docs on docs.getdbt.com.

Nodes, unique_id, and discovery

A node is a resource dbt tracks: models, sources, tests, seeds, snapshots, and more. Each node has a stable unique_id such as model.my_project.orders (package and name segments match dbt’s artifact format).

• dbt selection (dbt run --select ...) chooses what dbt executes on the next run. See Node selection syntax for upstream grammar—this site does not document full selector syntax.
• dbt-tools discover ranks resources in already-produced artifacts using tokens like type:model. Intent and parity with CLI/Web are described in Discovery parity.

After a dbt run, check readiness and find resources:

# After a dbt run, check readiness
npx @dbt-tools/cli status --dbt-target ./target --json

# Find resources; note unique_id in output
npx @dbt-tools/cli discover --dbt-target ./target "orders" --limit 5 --json

# Explain one node by unique_id
npx @dbt-tools/cli explain model.my_project.orders --dbt-target ./target --json

Example discover output shape (synthetic IDs only):

{
  "matches": [
    {
      "unique_id": "model.my_project.orders",
      "resource_type": "model",
      "name": "orders"
    }
  ]
}

Sources and freshness

• Sources appear in manifest.json as source.* nodes. Lineage and dependency commands treat them like other graph nodes.
• Freshness is configured and evaluated in dbt (dbt source freshness); results may appear in sources.json when present.
• dbt-tools may surface source metadata from artifacts; it does not run warehouse freshness checks.

See Sources on docs.getdbt.com for freshness configuration.

Parsing boundary

Artifact parsing and type definitions come from the external dbt-artifacts-parser npm package. This repository owns analysis, CLI, MCP, and web layers on top of that parser.

Safe examples

Documentation and public sites must use synthetic project names and metadata. Do not paste real warehouse credentials, customer data, or production manifests into examples.

Further reading

dbt Labs (artifact authority)

• dbt artifacts reference
• Build and view your docs
• Node selection syntax
• Sources

dbt-tools site

• New to dbt?
• Version compatibility
• Try with a sample project
• Operational intelligence