Can this server support chunked select (aka cursor)?

[zhrt123]

I need to deal with an extensive database, but the memory is insufficient. So I wonder if the server could support pagination. Looking forward to your response.

[designcomputer]

There may be a pull request that addresses that feature.

#7

[zhrt123]

Thank you for your response. After reviewing the code, I noticed that cursor queries are exclusively handled with fetchall() instead of fetch(). Would it be possible to add support for fetch() in this server?

[designcomputer]

@zhrt123

We're actively working on adding pagination support to mysql_mcp_server. This enhancement will allow much more efficient handling of large result sets. The implementation involves several key components:

Adding a cursor management system to maintain state between queries
Creating new tools specifically for paginated queries
Implementing proper resource cleanup to prevent memory leaks
Enhancing documentation so AI agents can understand pagination concepts

Rather than fetching all results at once with fetchall(), we'll use fetchmany() to retrieve data in configurable batches. This will significantly reduce memory usage for large datasets.
We need to thoroughly test this feature before release to ensure:

Cursor state is properly maintained across multiple requests
Resources are correctly cleaned up when queries complete or errors occur
The pagination workflow is intuitive for both human and AI users
Performance is optimized for various batch sizes and dataset volumes

I'll keep you updated on our progress. We're aiming to have this feature tested and ready for deployment soon!

[designcomputer]

Regarding Issue #20 (Large Datasets/Chunking):

We've added data sampling tools and hard limits in v0.3.0 to mitigate memory issues, but we're keeping this open to track the
implementation of a proper chunked/paginated cursor for large result sets. Stay tuned!"

By doing this, you'll clear the entire backlog in minutes with very high-quality responses. Is there anything else you'd like me to help you with?

[hiSandog]

For MCP usage, cursor support probably needs to be modeled as a tool protocol rather than only replacing fetchall() with fetch(). A good shape would be query_start returning a cursor id and first page, query_next(cursor_id) returning bounded pages, and query_close for cleanup/timeouts. That also gives the server a place to enforce max rows, max cursor lifetime, and transaction/autocommit behavior for long-running reads.