[finklen-amazon]

Fixes

Summary

Cost Explorer returns DataUnavailableException: Insufficient amount of historical data. in two cases: when there is too little history to forecast, and when the forecast filter as a whole matches no past costs. That covers a misspelled value, the wrong case, a value from another dimension, or a partial value used with STARTS_WITH, CONTAINS or CASE_INSENSITIVE. The service accepts those MatchOptions but does not apply them; the DimensionValues reference says MatchOptions applies only to cost category and anomaly subscription actions.

The shared DataUnavailableException hint already names an unmatched filter as one possible cause, but it maps to do_not_retry. The forecast notes described only the history case. So a caller with a mistyped filter was told not to retry and reported that no forecast was possible, while the same account forecasts fine with a correct filter.

Changes

• getCostForecast / getUsageForecast: when a filter was sent and the service returns DataUnavailableException, the response now has retry_strategy: modify_request and a forecast-specific hint: do not resend the same request; check the values with getDimensionValues (getTagsOrValues for tags, getCostCategories for cost categories) and retry with corrected values; if the values exist and can occur together, report that there is too little history. Without a filter, the shared do_not_retry guidance is unchanged.
• Forecast notes: filter values must match exactly, including case; those MatchOptions are accepted but not applied; a filter that matches no past costs returns the same error as too little history.

No tool, parameter or schema is renamed or removed.

User experience

Before: {"operation": "getCostForecast", "metric": "UNBLENDED_COST", "granularity": "MONTHLY", "filter": {"Dimensions": {"Key": "SERVICE", "Values": ["Amazon Elastic Compute Cloud - Compte"]}}} returned DataUnavailableException "Insufficient amount of historical data." with retry_strategy: do_not_retry, and an agent reported that the account cannot be forecast.

After: the same call returns the same error_type and message, with retry_strategy: modify_request and a hint to check the filter values first. Unfiltered calls are unchanged.

Testing

• Unit: new tests for the hint, the notes and the guards: an unfiltered call keeps do_not_retry, an empty filter ({}, '{}') is a validation_error without the hint, and other errors with a filter are unchanged. 246 tests in the two touched test files pass; ruff and pyright are clean.
• Live (read-only, against a real account, through the MCP server over stdio):
  • On the change, all claims passed: the notes as served; a valid SERVICE filter forecasts successfully (control); a one-letter typo, a STARTS_WITH partial value and a misspelled USAGE_TYPE_GROUP each return DataUnavailableException with modify_request and the new hint.
  • On the base, the same harness failed 4 of 5 (old notes; do_not_retry with the generic hint). Only the valid-filter control passed.
  • Independently re-checked with direct API calls: CONTAINS and CASE_INSENSITIVE are ignored (same error); a value from another dimension, a misspelled tag value and the wrong case match nothing; Not of a misspelled value and Or of a valid and a misspelled value both forecast normally.

Merge conflict

#4919 merged as a squash commit, so this branch, which was stacked on #4919's branch, conflicted with main in cost_explorer_operations.py and cost_explorer_tools.py: the forecast lines #4919 changed showed up as edits to the same lines this PR extends. I rebased this PR's single commit onto main, dropping #4919's commits, which are now upstream. The commit applied with no manual edits. #4919's period_note text and helper are kept as merged, and this PR's changed lines are identical to the reviewed revision. Unit tests (3252), ruff and pyright pass on the rebased branch.

After this PR was approved, #4936 merged and added its billing_view_arn tests at the end of tests/tools/test_cost_explorer_tools.py, where this PR adds its forecast-notes test, so the branch conflicted with main again. I rebased onto main and kept both: #4936's tests unchanged, followed by this PR's test. No source or tool text conflicted, and this PR's changed lines are identical to the approved revision. Unit tests (3442), ruff and pyright pass on the rebased branch.

Main then moved again and this branch conflicted a third time. This time I merged main into the branch (a merge commit, no force-push):

• cost_explorer_tools.py, getUsageForecast: fix(billing-cost-management-mcp-server): reject group_by where Cost Explorer has no GroupBy and document filter limits #4964 replaced Example 3 (group_by is now rejected on forecasts) while this PR extended the Notes line. I kept fix(billing-cost-management-mcp-server): reject group_by where Cost Explorer has no GroupBy and document filter limits #4964's Example 3 and this PR's Notes line.
• tests/tools/test_cost_explorer_tools.py: both sides added tests at the end of the file; I kept both.
• tests/tools/test_cost_explorer_operations.py (no text conflict): fix(billing-cost-management-mcp-server): reject an empty filter instead of dropping it #4962 now rejects an empty filter with a validation_error before any call. This PR's unfiltered guard had used {} and '{}' as "no filter", so it now uses None only, and a new test checks that an empty filter gets the validation_error and never the no-match hint.

The tool code of this PR is unchanged. The live check passes 5 of 5 on the merge, and the follow-up commits change only that empty-filter test. The touched tests, ruff and pyright pass. One test in test_recommendation_details_tools.py fails on main too, without this PR.

Checklist

If your change doesn't seem to apply, please leave them unchecked.

• 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

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 (7b517b7).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4940   +/-   ##
=======================================
  Coverage   93.79%   93.79%           
=======================================
  Files        1080     1080           
  Lines       94586    94597   +11     
  Branches    15402    15403    +1     
=======================================
+ Hits        88715    88726   +11     
  Misses       3498     3498           
  Partials     2373     2373           

☔ 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.