fix(bedrock-kb-retrieval): preserve non-ASCII text in tool results

[kimnamu]

Thank you for maintaining the AWS MCP servers.

Fixes #3820

Summary

QueryKnowledgeBases and ListKnowledgeBases serialized results with json.dumps(...) (default ensure_ascii=True), so non-ASCII KB content (CJK, accented, emoji) was returned as \uXXXX escapes — token bloat + unreadable payloads. This adds ensure_ascii=False to the two user/model-visible serialization points (the MCP transport is already UTF-8).

Changes

• knowledgebases/retrieval.py:113, server.py:128 — add ensure_ascii=False
• regression tests in test_retrieval.py + test_server.py
• ## [Unreleased] / ### Fixed CHANGELOG entry

Before / After

Item	Before	After
KB text 東京は日本の首都です。 in tool output	❌ 東京... (escaped)	✅ 東京は日本の首都です。
KB name 한국어 데이터 소스	❌ 한국...	✅ 한국어 데이터 소스
Output is valid JSON / round-trips via json.loads	✅	✅ (unchanged)
Public API / tool signatures / result ordering	—	✅ Unchanged
ASCII-only content	✅ identical bytes	✅ identical bytes

Tests

40 passed, ruff check clean. Reverting the source change (keeping the tests) makes the 2 new tests FAIL with assert '東京...' in '...\\u6771\\u4eac...', proving they catch the bug; restoring it → all pass.

---

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.

[kimnamu]

Status update (no action needed — just keeping this current):

• Re-verified today against main: both call sites still serialize with default ensure_ascii=True (knowledgebases/retrieval.py:113, server.py:128), so the issue in bedrock-kb-retrieval: QueryKnowledgeBases/ListKnowledgeBases return non-ASCII (CJK, emoji) as \uXXXX escapes #3820 still reproduces and this fix is still applicable.
• Branch merges cleanly with current main (no conflicts) and the full server test suite passes locally: 40 passed, ruff check clean.

Happy to rebase or adjust anything (e.g. drop the CHANGELOG entry) if preferred — thanks for your time reviewing!

[github-actions]

This pull request is now marked as stale because it hasn't seen activity for a while. Add a comment or it will be closed soon. If you wish to exclude this issue from being marked as stale, add the "backlog" label.

[kimnamu]

Re-verified after the stale bump: rebased locally onto current main (clean, no conflicts) and ran the full suite for bedrock-kb-retrieval-mcp-server — 40 passed, ruff check clean. Both call sites (retrieval.py:113, server.py:128) still default to ensure_ascii=True, so the fix is still needed and unchanged from the original PR.

Still happy to adjust anything (e.g. drop the CHANGELOG entry) if that unblocks a review — thanks for keeping this open.

[github-actions]

This pull request is now marked as stale because it hasn't seen activity for a while. Add a comment or it will be closed soon. If you wish to exclude this issue from being marked as stale, add the "backlog" label.

[kimnamu]

Still relevant as of today's main (ebcafaed):

• Both call sites still use the default ensure_ascii=True (knowledgebases/retrieval.py:113, server.py:128), so bedrock-kb-retrieval: QueryKnowledgeBases/ListKnowledgeBases return non-ASCII (CJK, emoji) as \uXXXX escapes #3820 reproduces unchanged.
• This branch merges cleanly into current main, and uv run --frozen pytest in src/bedrock-kb-retrieval-mcp-server gives 40 passed with ruff check / ruff format --check clean.

Two things that may help a decision:

1. The same one-line change was merged for a sibling server in fix(amazon-bedrock-agentcore-mcp-server): preserve non-ASCII text in browser_evaluate results #3995 (browser_evaluate, ensure_ascii=False), so the approach already has precedent in this repo.
2. fix(bedrock-kb-retrieval): preserve non-ASCII characters in tool output #4372 proposes the same change to the same two lines. No need to review both — if that one is easier to take, I'll close this and bedrock-kb-retrieval: QueryKnowledgeBases/ListKnowledgeBases return non-ASCII (CJK, emoji) as \uXXXX escapes #3820 can point there instead.

If this is only waiting on review bandwidth, the backlog label would keep the stale bot off it. Thanks for your time.