🐞 Bug report: list_spo_page LM tool doesn't return complete data

[Saurabh7019]

⭐ Priority

(Low)☹️ Something is a little off

📝 Describe the bug

The list_spo_page language model tool is returning incomplete results when listing SharePoint pages. In most cases, not all pages from the underlying CLI for M365 command are displayed, even though the CLI itself returns the full set of pages.

👣 Steps To Reproduce

1. Use a SharePoint site with multiple modern pages (e.g., create a new communication site using the OOB Department template).
2. Verify the total pages using the CLI for Microsoft 365 command: m365 spo page list --webUrl <site-url>
3. Ensure all pages are returned.
4. Next, use the #SharePointPageList LM tool to list the pages.
5. Compare the results with the CLI output

📜 Expected behavior

list_spo_page LM tool should return correct response. Number of pages, and then a summary.

📷 Screenshots

❓SharePoint Framework Toolkit version

4.16.4

❓Node.js version

24.12.0

🤔 Additional context

The issue seems to be the with the large JSON (especially CanvasContent1 field) that causes the entire response to exceed the token limits and the LM then truncates the response.

The solution could be to parse the JSON response from CLI command stdout and extract only relevant fields.

[Adam-it]

I had a similar issue in the CLI for Microsoft 365 MCP server that for the m365... list related commands it was cluttering the whole context window.
What I did there to solve the problem is to dynamically update the output mode to CSV for or list command types, as this is more efficient for array based responses for LLMs. For detail results, so for get command types I left JSON as this is better.

I would suggest similar approach here.
Also in CLI for M365 for text and CSV output mode we already trim the result to what think is minimal and most relevant

[nicodecleyre]

Can we open this up with the workaround that @Adam-it proposes? If this is a general issue with get and or list command types, then maybe we should change this generally on our side?

[Adam-it]

@nicodecleyre IMO yes I would perform a fix that for list command types only, changes the output mode from JSON to CSV.
so if we have an LM tool that is a wrapper around a list command type we should update to CSV
get command types do not return sooooo much data so JSON will be good enough.