[he-yufeng]

Summary

• normalize redirect_uris values at the OAuth client metadata boundary
• keep raw string, AnyUrl, and URL subtype inputs serializing the same way
• add a regression test for AnyHttpUrl registration followed by AnyUrl redirect validation

To verify

• .\.venv\Scripts\python.exe -m pytest tests\shared\test_auth.py -q
• .\.venv\Scripts\python.exe -m ruff check src\mcp\shared\auth.py tests\shared\test_auth.py
• .\.venv\Scripts\python.exe -m ruff format --check src\mcp\shared\auth.py tests\shared\test_auth.py
• git diff --check

Refs #2687

[he-yufeng]

Covered the additional collection inputs in 99b80b5.

The before-validator now normalizes list, tuple, set, and frozenset redirect URI inputs before Pydantic coerces the field, so URL subtype elements compare correctly after validation.

Validation:

uv run --frozen pytest tests/shared/test_auth.py -q
# 16 passed

uv run --frozen ruff check src/mcp/shared/auth.py tests/shared/test_auth.py
# All checks passed

uv run --frozen ruff format --check src/mcp/shared/auth.py tests/shared/test_auth.py
# 2 files already formatted

uv run --frozen pyright src/mcp/shared/auth.py tests/shared/test_auth.py
# 0 errors

python -m py_compile src\mcp\shared\auth.py tests\shared\test_auth.py

git diff --check

[he-yufeng]

Closing this older redirect URI normalization path in favor of #2808. The newer PR now includes the tuple/set/frozenset collection coverage from this branch, so review can continue on one green, current path.