[mattkelsey]

Fixes

Summary

Changes

ListRecommendationSummaries returns at most 1,000 groups, the ones with the highest estimated savings, and no nextToken for the rest. With group_by=AccountId in an organization where more than 1,000 accounts have recommendations, list_recommendation_summaries fetched every page and reported complete_dataset: true. Callers then presented the top 1,000 accounts as every account, leaving out accounts that had matching recommendations.

• A result that holds 1,000 groups is now marked complete_dataset: false, with has_more: false, no next_token, and a warning. The warning says groups may be missing and estimated_total_deduped_savings may cover only the returned groups. It also says to group by a dimension with fewer values for totals. It does not suggest a way to fetch the missing groups. total_recommendation_count is left out, as on any partial result.
• With TagKey: grouping, the warning adds that NoTagKey may be missing, or may also count recommendations whose tag value was left out.
• The tool description states the limit, and that neither a larger max_items nor a continuation can fetch past it. A continuation call can't count earlier pages, so its last page isn't marked; the description tells callers who page with max_items that 1,000 groups across all pages means the same.

Results under 1,000 groups are unchanged. A result of exactly 1,000 groups can still be complete, so the warning says groups may be missing.

User experience

Before: a 1,000-group summaries result looked complete, so the agent answered "which accounts have X?" with exactly 1,000 accounts and no caveat.

After: the same result leads with complete_dataset: false and a warning. The agent says the list covers the 1,000 groups with the most savings, and takes totals from a coarser grouping such as Region.

Testing

• 9 new unit tests. 5 fail on the base and pass with the change: a 1,000-group result in one page, 1,000 groups across two pages, max_items=1000, the TagKey: warning, and the served tool description. 4 are regression guards that pass on both: 999 groups stay complete with total_recommendation_count, 999 TagKey: groups stay complete, a continuation is not flagged, and a max_items partial with a token keeps its own warning.
• Full suite: 4,053 passed. ruff check, ruff format --check and pyright are clean.
• Served description, checked over stdio through list_tools: the 5 new statements are present with the change and absent on the base, and the description suggests no way to fetch the missing groups.
• I checked the limit, the missing token, and what the total covers against the service's behavior. The limit itself couldn't be reproduced in a test account, which has far fewer than 1,000 groups.
• This touches the same description bullets as fix(billing-cost-management-mcp-server)!: request COH savings percentage, coerce boolean filters, fold summaries group_by #4906, so whichever merges second will need a small rebase.

Checklist

• I have reviewed the contributing guidelines
• I have performed a self-review of this change
• Changes have been tested
• Changes are documented

Is this a breaking change? (Y/N) N

RFC issue number:

Checklist:

• Migration process documented
• Implement warnings (if it can live side by side)

Acknowledgment

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of the project license.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.79%. Comparing base (c6951b9) to head (3740adf).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4973      +/-   ##
==========================================
- Coverage   93.79%   93.79%   -0.01%     
==========================================
  Files        1080     1080              
  Lines       94586    94595       +9     
  Branches    15402    15404       +2     
==========================================
+ Hits        88715    88723       +8     
  Misses       3498     3498              
- Partials     2373     2374       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mattkelsey]

I don't love hard-coding this, but there is not a source we can consume it from.

In botocore model there is max page size, but that is not the same as max groups. No model exists for this

[pputra]

remove "for rows and counts, use list_recommendations with the same filters (its savings are not de-duplicated)"

[mattkelsey]

ack

[mattkelsey]

is there a different alternative we can give it?

[pputra]

leaving some comments

[mattkelsey]

Had to rebase sorry @wangyuhere

[pputra]

no permission to approve, but LGTM