Airbyte's 700+ connectors, as a Python library.

PyAirbyte is an open-source Python library that runs Airbyte connectors in your code. Pull data from APIs, databases, and files into DuckDB, Postgres, Snowflake, or BigQuery, then work with it in pandas, SQL, or your AI framework.

BASH
pip install airbyte

Python 3.10 to 3.12. Using uv? Run uv add airbyte.

Data ingestion in a few lines of Python

Pick a connector, select streams, and read. PyAirbyte installs the connector, tracks incremental state, and caches the records locally so you can query them right away.

35,000+

custom connectors built

25,000+

companies run Airbyte

Open source

Read the code, file issues, and send pull requests on GitHub

Read GitHub issues into a local DuckDB cache, then hand them to pandas. Run it again and incremental streams fetch only what's new.

PYTHON
import airbyte as ab

source = ab.get_source(
    "source-github",
    config={
        "repositories": ["airbytehq/PyAirbyte"],
        "credentials": {
            "personal_access_token": ab.get_secret("GITHUB_PERSONAL_ACCESS_TOKEN"),
        },
    },
)
source.check()
source.select_streams(["issues"])

result = source.read()  # writes to a local DuckDB cache by default
issues = result["issues"].to_pandas()

How does PyAirbyte work?

PyAirbyte runs the same connectors that power Airbyte Cloud and Airbyte Core (open source), on your laptop, in a notebook, or wherever your Python code runs. It's built for Python developers who want to build data pipelines in their own code.

Data flow diagram: a source connector reads from an API, database, or file, PyAirbyte loads the records into a SQL cache or an Airbyte destination, and your code queries them with pandas, SQL, or AI frameworks.

Source connector

APIs, databases, files

PyAirbyte

Runs the connector from your code.

SQL cache

DuckDB, Postgres, Snowflake, BigQuery, or an Airbyte destination

Your code

pandas, SQL, AI frameworks

What you can build with PyAirbyte

  • Prototype a pipeline in a notebook

    Test a source in Jupyter or Google Colab before you commit to a connection. Read a few streams, check record counts, and shape the data in pandas.

  • Replace custom ETL scripts

    Swap one-off API scripts for Airbyte connectors that already handle pagination, rate limits, and incremental state.

  • Run pipelines as code

    Apply data engineering best practices to your pipelines. Keep the logic in version control, review it in pull requests, test it in CI, and run it from Airflow, Dagster, or any orchestrator that runs Python.

  • Feed analysis and AI apps

    Load records into pandas or query them with SQL. Then pass them to LangChain, LlamaIndex, or whatever framework runs your LLM app.

  • Build and test connectors

    Run a connector from a local directory or a Connector Builder YAML manifest, and check it against real data before you publish it.

  • Explore data without a warehouse

    Read a source into the local DuckDB cache and query it with SQL on your laptop. No destination to configure, and no warehouse account to wait on.

PyAirbyte MCP: run Airbyte from your AI client

The Airbyte MCP server ships in PyAirbyte. The same open-source package runs the hosted server at mcp.airbyte.com and the one you run yourself.Ask Claude, ChatGPT, or Cursor to build a pipeline, find out why a sync failed, or read a job log. The agent works with your Airbyte permissions.The Airbyte MCP is free on every plan, including Airbyte Core.
  • Airbyte Cloud and Airbyte Flex

    Connect your MCP client to https://mcp.airbyte.com/mcp and sign in with your Airbyte account.

  • Airbyte Core (open source)

    Run the server yourself with uvx --from=airbyte@latest airbyte-mcp and point it at your instance.

What PyAirbyte handles for you

  • 700+ connectors

    Use the same connector catalog as Airbyte Cloud. Python and YAML connectors run natively, and the rest run in Docker automatically.

  • SQL caches

    Records land in a local DuckDB file by default. Swap in Postgres, Snowflake, or BigQuery by passing a different cache to read().

  • Destinations

    Call destination.write(source) to send data to an Airbyte destination, with incremental state carried between runs.

  • Incremental syncs

    PyAirbyte stores stream state alongside your data, so each run picks up where the last one stopped.

  • Isolated installs

    Each connector installs into its own virtual environment with uv. Set a different Python version per connector with use_python.

  • Airbyte Cloud from Python

    Deploy sources, destinations, and connections, set schedules, run syncs, and read results from Snowflake and BigQuery destinations.

Start with the library. Run production on Airbyte.

PyAirbyte doesn't schedule syncs, send alerts, or manage access. It's a library for code you own. When a pipeline needs those things, move it to Airbyte.PyAirbyte is free. Compare Airbyte Cloud plans on the pricing page.
  • PyAirbyte

    Free and open source. Connectors in your code, local caches, and full control over where it runs.

    BASH
    pip install airbyte
  • Airbyte Cloud

    Scheduling, monitoring, and alerting, with SSO, RBAC, and audit logging on the plans that include them. Try it free for 14 days.

  • Airbyte Flex

    Airbyte's data plane runs in your cloud, VPC, or on-prem environment, and Airbyte manages the control plane. Your data never touches our servers.

Frequently asked questions

PyAirbyte is an open-source Python library that runs Airbyte connectors in your code. Install it with pip install airbyte, configure a source, and read records into a local cache or write them to an Airbyte destination. It gives Python developers access to Airbyte's 700+ connectors without deploying an Airbyte instance. PyAirbyte supports Python 3.10 to 3.12.

No. PyAirbyte is a library for running connectors from Python code. It doesn't schedule syncs, monitor them, send alerts, or manage access. Use it to prototype, script, and test pipelines, then deploy them to Airbyte Cloud, or to Airbyte Flex if your data plane has to run in your own environment. PyAirbyte can deploy sources, destinations, and connections to Airbyte for you.

The cache is a SQL database where PyAirbyte stores the records it reads. DuckDB is the default and needs no setup, and you can also use Postgres, Snowflake, or BigQuery. It's called a cache to keep it separate from Airbyte destinations. PyAirbyte can also write directly to Airbyte destinations with destination.write(), which is the better fit for targets that aren't SQL databases.

Yes. PyAirbyte covers the extract and load steps: it reads from a source and writes to a cache or an Airbyte destination. Transform the data with pandas or SQL, in the same script or in a later job. That covers both traditional ETL scripts and ELT pipelines that load first and transform in the warehouse.

Millions of rows, yes, and incremental syncs keep later runs small because they only move new data. For terabyte-scale or always-on workloads, run the connection on Airbyte Cloud or Airbyte Flex, which add scheduling, monitoring, and alerting.

Common uses include prototyping a pipeline in a notebook before you deploy it, replacing custom API scripts with Airbyte connectors, running pipelines as code in CI or an orchestrator, loading data into pandas or AI frameworks like LangChain and LlamaIndex, developing or testing your own connectors, and exploring a source in a local DuckDB cache before you pick a warehouse.

PyAirbyte reads each stream against the schema the connector declares. By default, it drops fields that aren't in that declared schema, which matches how PyAirbyte caches and most Airbyte destinations behave.

In the cache, in an internal table named _airbyte_state. Each run reads the saved state and fetches only the records that changed since the last run. When you write to a destination, PyAirbyte checks the cache for state first, then the destination, and runs a full refresh if it finds none.

Yes. The airbyte.cloud module connects to a workspace with client credentials. From there, you can deploy sources, destinations, and connections, set schedules, run syncs, and check job status. For Snowflake and BigQuery destinations, you can read sync results straight into Python. It works with Airbyte Cloud, Airbyte Flex workspaces, and Airbyte Core.

Yes. PyAirbyte is a regular Python package, so any orchestrator that runs Python can run it, including Airflow and Dagster. Run reads as tasks when you want full control, or use PyAirbyte to trigger syncs on Airbyte Cloud connections and let Airbyte do the work.

Yes. Point PyAirbyte at a connector in a local directory, a connector on your PATH, or a Connector Builder YAML manifest, then read real data from it in a notebook or a test suite. It's a quick way to check streams and record shapes before you publish a connector.

The Airbyte MCP server ships in PyAirbyte. The hosted server at mcp.airbyte.com runs from the same open-source package, and Airbyte Core users run it themselves with uvx --from=airbyte@latest airbyte-mcp. Use it to build pipelines, troubleshoot failed syncs, and read job logs from Claude, ChatGPT, Cursor, or another MCP client.