Status: Public Preview — behaviour may change.
What it does
An open-source set of instructions that lets a supported coding agent carry out Dataverse development and admin tasks from natural-language prompts. You describe the outcome; the agent picks the right skill and the right underlying tool. It doesn't replace the Dataverse tooling — it tells the agent how to drive it.
Key facts
- Sits on top of the existing tools rather than replacing them: the Dataverse CLI, Dataverse MCP server, Dataverse SDK for Python, Power Platform CLI and the Web API.
- The agent loads an overview skill for context, then pulls in specialised skills (metadata, data, query, solution) per request — one prompt can chain several.
- Tool selection is guided by operation and data volume: MCP server for interactive queries and small metadata/record work, CLI for auth and headless data ops, Python SDK for bulk and scripted schema work, Power Platform CLI for solution ALM and environment/role ops, Web API for anything not covered by a managed tool.
- Supported agents: GitHub Copilot, Claude Code, Cursor, Codex.
- Source lives in the microsoft/Dataverse-skills GitHub repository.
- Needs a Dataverse environment (a Developer Plan environment is fine for non-production) and an account with the privileges for whatever you ask it to do.
When to use / skip
Reach for it when a task spans several tools — build a data model, load records, then move the solution between environments — and you'd rather not hand-write the syntax for each. Skip it for a one-line Web API call you already know; the plugin earns its keep on multi-step, multi-tool work.
Configuration decisions
- Which coding agent to standardise on, since the plugin supports several and your team's existing tooling will steer the choice.
- Which local tools to install and permit — the agent can only drive what's present and authorised.
- How to scope environment access for the account the agent runs as; it will act with those privileges.
Gotchas
- It's preview and open-source guidance, not a supported product surface — treat generated changes as proposals, not done deals.
- The plugin applies confirmation steps before changes that touch an environment, solution, data or security, but you still own reviewing the proposed change and confirming the target environment. An agent pointed at the wrong environment will happily do the right operation in the wrong place.
Consultant notes
- The obvious risk isn't the agent getting Dataverse syntax wrong — it's the agent operating against production because nobody scoped its environment access. Give it a non-production environment and least-privilege credentials before anyone gets comfortable.
- Because it orchestrates five different tools, the value shows up in ALM-heavy work; teams still doing everything by hand in the maker portal won't feel it. Pitch it to developer teams already living in a coding agent, not to citizen makers.
- Preview means the skill set and behaviour will shift — don't build a delivery process that assumes today's skills are stable, and keep a human confirmation gate on anything that mutates an environment.
Preview feature — revisit when it moves toward general availability