fix(ci): Fix mypy type errors across 6 files

[shin-bot-litellm]

Fix

Resolves mypy type errors in 6 files:

• vertex_ai/gemini/transformation.py — TypedDict assignment via explicit dict cast
• mcp_server/server.py — convert MutableMapping scope to dict
• pass_through_endpoints.py — simplify custom_body logic for type narrowing
• vector_store_endpoints/endpoints.py — add Any annotation for dynamic hook return
• responses/transformation.py — use dict() for Reasoning, setattr for dynamic field
• zscaler_ai_guard.py — add assert for api_base None check

No pyproject.toml changes.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Building	Preview, Comment	Feb 14, 2026 2:13am

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Overview

Greptile Summary

This PR resolves mypy type errors across 6 files using a variety of approaches: explicit dict casts, type: ignore comments, isinstance narrowing, Any annotations, setattr for dynamic fields, and an assert for None checks. All changes are type-annotation-level fixes with no intended behavioral changes. Most fixes are clean and appropriate, though the assert in zscaler_ai_guard.py could be problematic if Python is ever run with the -O flag (which strips assertions).

• vertex_ai/gemini/transformation.py: Uses data_dict: dict = data # type: ignore to allow arbitrary key assignment into a TypedDict
• mcp_server/server.py: Converts ASGI MutableMapping scope to dict() for type compatibility
• pass_through_endpoints.py: Simplifies ternary into if/elif for better type narrowing — behavior unchanged
• vector_store_endpoints/endpoints.py: Annotates hook return as Any to allow subclass method calls
• responses/transformation.py: Wraps Reasoning() in dict() and uses setattr for dynamic field — both functionally equivalent
• zscaler_ai_guard.py: Adds assert api_base is not None — correct at runtime but fragile under -O flag

Confidence Score: 4/5

• This PR is safe to merge — all changes are type-level fixes with no behavioral changes, though one assert-based fix has minor fragility.
• Score of 4 reflects that all 6 files contain purely type-annotation fixes with no behavioral changes. The only concern is the assert pattern in zscaler_ai_guard.py which, while correct in normal execution, could fail silently under Python's -O flag. The remaining 5 changes are clean and safe.
• Pay close attention to litellm/types/proxy/guardrails/guardrail_hooks/zscaler_ai_guard.py — the assert pattern is fragile in optimized Python execution.

Important Files Changed

Filename	Overview
litellm/llms/vertex_ai/gemini/transformation.py	Introduces a data_dict: dict = data alias with type: ignore to work around mypy's TypedDict assignment restrictions when merging extra_body. Functionally equivalent to the original code.
litellm/proxy/_experimental/mcp_server/server.py	Wraps ASGI scope (MutableMapping) in dict() to satisfy mypy's Dict type expectation for maybe_build_debug_headers. Safe and correct conversion.
litellm/proxy/pass_through_endpoints/pass_through_endpoints.py	Simplifies final_custom_body logic using elif isinstance(custom_body_data, dict) instead of a ternary expression. Behavior is identical to the original code; improves type narrowing for mypy.
litellm/proxy/vector_store_endpoints/endpoints.py	Adds Any type annotation to managed_vector_stores to suppress mypy errors when calling subclass-specific methods on a CustomLogger return type. Import of Any added.
litellm/responses/litellm_completion_transformation/transformation.py	Replaces Reasoning() with dict(Reasoning()) and direct attribute assignment with setattr() to resolve mypy errors. Both changes are functionally equivalent given the model's extra="allow" config.
litellm/types/proxy/guardrails/guardrail_hooks/zscaler_ai_guard.py	Adds assert api_base is not None for mypy type narrowing. While the value is always set via env default, assert can be stripped with -O flag, making this fragile in production.

Flowchart

flowchart TD
    A[mypy type check] --> B{Type error in file?}
    B -->|vertex_ai/transformation.py| C["TypedDict assignment\n→ dict alias + type: ignore"]
    B -->|mcp_server/server.py| D["MutableMapping scope\n→ dict(scope)"]
    B -->|pass_through_endpoints.py| E["Ternary expression\n→ if/elif narrowing"]
    B -->|vector_store_endpoints.py| F["CustomLogger return\n→ Any annotation"]
    B -->|responses/transformation.py| G["Reasoning TypedDict\n→ dict(Reasoning())\nDynamic field\n→ setattr()"]
    B -->|zscaler_ai_guard.py| H["Optional[str] usage\n→ assert not None"]
    C --> I[mypy passes]
    D --> I
    E --> I
    F --> I
    G --> I
    H --> I

Loading

_{Last reviewed commit: 29cf97d}

[greptile-apps]

_{6 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

assert can be stripped in production

Using assert for a runtime invariant is fragile — Python's -O (optimize) flag disables all assert statements. If the proxy is ever run with python -O, this assertion will be silently skipped and api_base could be None, causing an AttributeError on line 110. A safer pattern for a runtime check would be:

Suggested change

	assert api_base is not None # always set via env default above
	if api_base is None: # always set via env default above
	raise ValueError("api_base must not be None")

[greptile-apps]

type: ignore hides potential misuse

The # type: ignore[assignment] silences mypy but doesn't actually fix the underlying issue — RequestBody is a TypedDict with a fixed set of keys, and assigning arbitrary extra_body keys into it via data_dict[k] = v (line 541) would still be incorrect from a type-safety perspective. The cast to dict is just telling mypy to look away. A cleaner approach would be to use cast(dict, data) which is more explicit, or restructure to avoid the mismatch entirely. That said, this is a minor style concern and the runtime behavior is correct since TypedDicts are regular dicts at runtime.

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}