Status: Public Preview — behaviour may change.

What it does

Lets Python code talk to Dataverse directly, without any .NET or C#. You install a PyPI package (PowerPlatform-Dataverse-Client), authenticate with an Azure identity credential, and get CRUD, bulk operations, queries and table metadata management through Python syntax. It's aimed squarely at data scientists and automation authors who live in Pandas and notebooks rather than Visual Studio.

Key facts

• Package is PowerPlatform-Dataverse-Client from PyPI; source is on GitHub (microsoft/PowerPlatform-DataverseClient-Python) under an open-source licence.
• Needs Python 3.10+, Pandas 2.0.0+, and network access to pypi.org.
• Authentication is via Azure Identity credentials that expose get_token() — you bring your own OAuth setup, the SDK doesn't do interactive sign-in for you.
• Bulk operations map to Dataverse's native CreateMultiple / UpdateMultiple / UpsertMultiple / BulkDelete, so you get the same server-side performance and transactional changesets you'd get from the .NET SDK.
• CRUD wrappers can return Pandas DataFrames and Series directly, which is the whole point for analytics work.
• Covers table and column metadata CUD (create/update/delete custom tables and columns, optional solution association), a fluent QueryBuilder, read-only direct-SQL SELECTs with paging, file-column upload with chunking, automatic retries on throttling, and OptionSet-to-enum mapping.
• Structured exception hierarchy with retry guidance, plus opt-in HTTP diagnostics logging that redacts the authorization header.

When to use / skip

Use it when the consumer is Python — data science, ML feature pipelines, notebook-driven automation, or agentic workflows that already sit in the Python ecosystem. Skip it for anything that belongs in-platform (plug-ins, low-code flows) or where you'd otherwise use the .NET SDK or Web API from a supported server-side context. It's preview, so keep it out of anything you can't afford to rework.

Configuration decisions

• How you obtain the token: which Azure Identity credential type fits your host (managed identity, service principal, interactive) — the SDK assumes you've sorted this out.
• Install from the PyPI package or from GitHub source, not both — pick one per environment.
• Whether to pull DataFrames or raw records back; DataFrames are convenient but carry memory cost on large result sets.
• Whether to enable HTTP diagnostics logging, and where those log files land, given they capture request/response detail.

Gotchas

• It's Public Preview — no production SLA, and the API surface can shift between releases. Pin your package version.
• Direct-SQL support is read-only SELECT with paging only; don't expect to write through it.
• It leans on Pandas as a hard dependency (2.0.0+), so it pulls a real dependency footprint into environments that might otherwise be lean.
• Authentication is deliberately your problem — there's no built-in device-code or browser flow, so a misconfigured credential is the usual first-run failure.

Consultant notes

• The pitch is "no .NET needed", and that lands well with data teams — but the people who'll actually run this often aren't Power Platform admins, so scope who sets up the app registration and permissions early, or first-run auth stalls.
• Preview means preview: don't let a customer wire this into a scheduled production job on the strength of a demo. Two or three releases from now the exception types or method signatures may not match your code.
• Bulk operations are the reason to reach for it over ad-hoc Web API calls from requests — if someone's hand-rolling paged HTTP loops in Python, this replaces that with the native multiple-operation messages and proper changesets.
• Watch the Pandas version pin in shared environments; a notebook estate on older Pandas will need bumping before the SDK installs cleanly.

---

Preview feature — revisit when it reaches general availability, and re-check the API surface after each SDK release.