dcf-core 0.1.5

YAML-driven data ingestion framework. Define collectors, dcf handles the rest.

dcf

YAML-driven data ingestion. Define a collector — dcf fetches, projects, and writes it to your warehouse.

uvx --from dcf-core dcf init

---

How it works

1. Define a collector in YAML — source, schema, cadence
2. Run it with dcf run
3. Data lands in your local warehouse (Parquet + DuckDB) or a GCS-backed lake

Sources: HTTP REST/CSV APIs · Python functions · Google Pub/Sub
Write strategies: incremental (upsert) · append · full_refresh
Deploy targets: local (Docker + Airflow) or GCP (Cloud Composer + Dataflow)
Claude integration: dcf mcp setup-desktop registers an MCP server so Claude can write and run collectors on your behalf.

---

Example

name: dcf_commits
namespace: github

source:
  type: http
  url: https://api.github.com/repos/zephschafer/dcf/commits
  method: GET
  params:
    - name: per_page
      type: integer
      value: 100
  schema:
    columns:
      - {name: sha,          path: sha,                type: string}
      - {name: author,       path: commit.author.name, type: string}
      - {name: message,      path: commit.message,     type: string}
      - {name: committed_at, path: commit.author.date, type: timestamp}

cadence:
  strategy: incremental
  primary_key: sha

deployment:
  schedule: "0 8 * * *"

uv run dcf run dcf_commits
uv run dcf query 'SELECT * FROM github.dcf_commits LIMIT 5'

---

Install

pip install dcf-core

The CLI command is dcf.

---

Quickstart

mkdir my-project && cd my-project
uvx --from dcf-core dcf init
uv sync
uv run dcf run dcf_commits
uv run dcf query 'SELECT * FROM github.dcf_commits'

dcf init creates pyproject.toml, project.yml, .gitignore, collectors/, and an example collector.

---

Contributing

git clone https://github.com/zephschafer/dcf && cd dcf && uv sync

Point a local project at your checkout:

[tool.uv.sources]
dcf-core = { path = "../dcf", editable = true }

To verify changes:

uv run dcf run dcf_commits
uv run dcf query 'SELECT * FROM github.dcf_commits'

Releasing: bump version in pyproject.toml and push to main — GitHub Actions publishes to PyPI automatically.

---

Package structure

dcf/
├── cli.py              Entry point (Typer)
├── config/
│   ├── models.py       Pydantic models for collector YAML
│   └── loader.py       YAML loading + env var resolution
├── engine/
│   ├── runner.py       Outer loop (iterate → fetch → project → write)
│   ├── fetcher.py      HTTP and Python source fetchers
│   ├── iterator.py     Date range and categorical iteration
│   ├── projector.py    Schema projection and path extraction
│   └── transforms.py   Column transforms
├── writer/
│   └── iceberg.py      Write strategies (incremental / append / full_refresh)
└── gcp/                GCP auth, provisioning, Terraform wrappers