Dataverse MCP Server connector in Copilot Studio lacking features pointing to specific tables

Question

Since the release of Dataverse MCP server in Copilot Studio, we have been testing it. The result is far better than Dataverse Knowledge base. But we can't rollout this to production because it's still in preview and the result is inconsistent (sometime it's awesome, other time it's showing failure, no data available).

We have analyzed the issue further -

1. We have 1M records in Dataverse table
2. All the instruction and tables/columns schema have been provided in the instruction
3. Sometimes the sql query generated from the MCP is precise. But other time, the query is running into error even after multiple times.
4. It seems that the error is due to not having proper tables schema details. ideally it should use desc table which its not doing.

I would suggest to the Product team have a place holder(similar to description) in the Dataverse MCT Server tool connector to have a table schema section(where we can provide the tables/columns schema with details) or a table selection area (where we can select only the tables we need. This would help the tool to minimize the error.

Please let me know if anyone can help with this.

Answer 1

Hi Palash,

You’ve raised a very relevant point — the Dataverse MCP Server connector in Copilot Studio is still in preview, and its current behavior is expected to be somewhat inconsistent, especially with large datasets or complex table structures. The connector doesn’t yet fully support granular schema mapping or table-level context handling, which is why you’re seeing varying SQL generation accuracy.

Right now, the tool relies on contextual instructions rather than dynamic schema introspection (like DESC TABLE). This means the MCP model sometimes fails to interpret or recall schema details consistently, particularly when working with large data volumes such as 1M+ records.

Your suggestion to include a table schema definition or table selection option inside the connector setup is valid and aligns with several internal feedback requests already raised during preview. Providing schema metadata directly to the MCP would improve consistency in query generation and reduce “no data” or “query error” outcomes.

At this stage, the best workaround is:

Keep schema instructions concise and provide only the most used tables and columns.

Where possible, split the data into smaller logical entities and test MCP responses per table.

Log errors or correlation IDs for inconsistent responses and share them through the Copilot feedback channel or your Microsoft support contact — this helps engineering validate schema recognition issues.

This feedback should be logged as a Feature Ask for the product team since it’s a missing capability, not a defect in an existing function.

Relevant Microsoft Documentation:

1. Connect to Dataverse with model context protocol (preview)
2. Feedback in Copilot Studio
3. Create a new Model Context Protocol (MCP) server

Thank you.

Karan Shewale.

If this response resolves your issue, please Accept the answer and, if helpful, click the “Upvote” button. Your feedback helps us improve and assist others more effectively.

Answer 2

It appears that you are experiencing issues with the Dataverse MCP Server connector in Copilot Studio, particularly regarding the handling of table schemas and the generation of SQL queries. Since this feature is still in preview, it is not uncommon to encounter inconsistencies and errors, especially with large datasets like your 1M records.

The suggestion to include a placeholder for table schema details or a table selection area in the Dataverse MCP Server tool connector is a valid one. This could potentially help reduce errors by providing clearer context for the queries being generated. However, as this is a preview feature, feedback like yours is crucial for the product team to consider improvements.

For now, it may be beneficial to monitor the updates from Microsoft regarding the MCP server and participate in any feedback channels they provide, as they may address these issues in future releases.

---

References:

• Connect to Dataverse using an MCP Server with Claude or GitHub Copilot in VS Code
• Connect to Dataverse with model context protocol (preview)