[ChiragAgg5k]

What does this PR do?

Follow-up to #170, the same fix as appwrite/sdk-generator#2222 for the CLI.

When the SDK can't parse a response, the error returns the raw payload, and #170 redacts it by field name: every name an appwrite_console model marks format: password. That set includes code, from MFAChallengeSecret.code, the one-time MFA code. So an unparsed country, language or locale payload would have its ordinary code hidden.

_secret_names now leaves out any name that is also a non-secret string (or a list or optional of one) on another SDK model. Against appwrite-console 0.16.0, that drops only code. The other 30 names stay, including secret, password and uri.

Typed results aren't affected: they're redacted per model field, so MFAChallengeSecret.code is still redacted there.

Test Plan

• Extended test_format_appwrite_error_returns_the_unhydrated_payload with an unparsed country payload; "code": "DE" comes back untouched.
• ruff, black --check and pyright (0 errors) pass.
• Unit and e2e suites pass.

[appwrite]

🟡 Tier B · Needs changes before merging

An SDK parse failure on an MFA response can now expose its one-time code in the raw error payload.

This change excludes SDK field names that also appear as ordinary string fields from name-based redaction of unparsed responses. It adds a country-code regression case to the Appwrite error-formatting test.

Verdict	New comments	Fixed	Still open
🛑 Changes requested	1	0	0

	Finding	Where
🟠	Keep MFA code redacted in untyped responses	src/mcp_server_appwrite/redaction.py:131

Fix with agent prompt

### Issue 1
src/mcp_server_appwrite/redaction.py:131
**Keep MFA `code` redacted in untyped responses**

`_format_response_parse_error` uses `redact_payload` for every unhydrated response, so removing `code` here disables its masking globally, not just for Country. If an MFA challenge response fails SDK hydration, its password-formatted one-time code is serialized verbatim into the MCP error result and may be retained in the conversation transcript.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

📂 Walkthrough · 2

File	Change
src/mcp_server_appwrite/redaction.py	Collects ordinary text field names and subtracts them from the secret-name set.
tests/unit/test_server.py	Checks that an unparsed country payload retains its ordinary code.

---

_{Reviewed 0a8a4b9 · Details · Comment @appwrite review to re-run, or mention @appwrite with a question.}

[appwrite]

🟡 Tier B · 1 blocking finding to address. Summary

[appwrite]

Keep MFA code redacted in untyped responses

_format_response_parse_error uses redact_payload for every unhydrated response, so removing code here disables its masking globally, not just for Country. If an MFA challenge response fails SDK hydration, its password-formatted one-time code is serialized verbatim into the MCP error result and may be retained in the conversation transcript.

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/mcp_server_appwrite/redaction.py
Line: 131

Comment:
**Keep MFA `code` redacted in untyped responses**

`_format_response_parse_error` uses `redact_payload` for every unhydrated response, so removing `code` here disables its masking globally, not just for Country. If an MFA challenge response fails SDK hydration, its password-formatted one-time code is serialized verbatim into the MCP error result and may be retained in the conversation transcript.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

_{🟠 Major · security · Reply if this doesn't apply.}

[ChiragAgg5k]

Closing: the review is right. The by-name fallback only runs when the SDK can't parse a response, and over-masking a country's code there is harmless, while leaking an MFA challenge code isn't. #170's behaviour (redact every name any model marks secret) is the safer trade-off for that rare path. The CLI case (appwrite/sdk-generator#2222) differs: it redacts every normal response by name, so it has to leave code alone.