Implement API for OpenAI chat history

Author: Johan Broberg

CLAUDE.md

@@ -168,12 +168,45 @@ Place it before imports with one blank line after.
 
 ### Python Conventions
 
-- Type hints preferred (Pydantic models heavily used)
+- Type hints required on all function parameters and return types
 - Async/await patterns for I/O operations
 - Use explicit `None` checks: `if x is not None:` not `if x:`
 - Local imports should be moved to top of file
 - Return defensive copies of mutable data to protect singletons
 
+### Type Hints - NEVER Use `Any`
+
+**CRITICAL: Never use `typing.Any` in this codebase.** Using `Any` defeats the purpose of type checking and can hide bugs. Instead:
+
+1. **Use actual types from external SDKs** - When integrating with external libraries (OpenAI, LangChain, etc.), import and use their actual types:
+   ```python
+   from agents.memory import Session
+   from agents.items import TResponseInputItem
+
+   async def send_chat_history_async(self, session: Session) -> OperationResult:
+       ...
+   ```
+
+2. **Use `Union` for known possible types**:
+   ```python
+   from typing import Union
+   MessageType = Union[UserMessage, AssistantMessage, SystemMessage, Dict[str, object]]
+   ```
+
+3. **Use `object` for truly unknown types** that you only pass through:
+   ```python
+   def log_item(item: object) -> None: ...
+   ```
+
+4. **Use `Protocol` only as a last resort** - If external types cannot be found or imported, define a Protocol. However, **confirm with the developer first** before proceeding with this approach, as it may indicate a missing dependency or incorrect understanding of the external API.
+
+**Why this matters:**
+- `Any` disables all type checking for that variable
+- Bugs that type checkers would catch go unnoticed
+- Code readability suffers - developers don't know what types to expect
+- Using actual SDK types provides better IDE support and ensures compatibility
+- This applies to both production code AND test files
+
 ## CI/CD
 
 The `.github/workflows/ci.yml` pipeline:

docs/prd/openai-send-chat-history-api.md

@@ -226,6 +226,52 @@ The API SHALL extract text content following this priority:
 | Type hints | Complete type annotations (PEP 484) |
 | Pydantic version | ≥2.0 (for model validation) |
 
+### 5.1.1 Type Hints - NEVER Use `Any`
+
+**CRITICAL**: The use of `typing.Any` is **strictly forbidden** in this codebase. Using `Any` defeats the purpose of type checking and can hide bugs.
+
+**Required alternatives (in order of preference):**
+
+| Instead of `Any` | Use |
+|------------------|-----|
+| External SDK types | Import and use actual types from the SDK (e.g., `Session`, `TResponseInputItem`) |
+| Multiple known types | `Union[Type1, Type2, ...]` |
+| Pass-through data | `object` |
+| Dictionary values | `Dict[str, object]` or specific types |
+| Unknown external types (last resort) | `Protocol` - but confirm with developer first |
+
+**Preferred approach - Use actual SDK types:**
+
+```python
+from agents.memory import Session
+from agents.items import TResponseInputItem
+
+async def send_chat_history_async(
+    self,
+    turn_context: TurnContext,
+    session: Session,  # Use actual SDK type
+) -> OperationResult:
+    ...
+
+async def send_chat_history_messages_async(
+    self,
+    turn_context: TurnContext,
+    messages: List[TResponseInputItem],  # Use actual SDK type
+) -> OperationResult:
+    ...
+```
+
+**Why actual SDK types are preferred:**
+- Better IDE support (autocomplete, type checking)
+- Ensures compatibility with the external SDK
+- Less maintenance burden (no custom protocols to keep in sync)
+- Clearer intent for developers reading the code
+
+**When to use Protocol (last resort only):**
+If external types cannot be found or imported, a `Protocol` may be defined. However, this should be rare and requires confirmation with the developer before proceeding, as it may indicate a missing dependency or incorrect understanding of the external API.
+
+This requirement applies to both production code AND test files.
+
 ### 5.2 OpenAI SDK Compatibility
 
 | Requirement | Specification |

libraries/microsoft-agents-a365-tooling-extensions-openai/microsoft_agents_a365/tooling/extensions/openai/__init__.py

@@ -2,10 +2,21 @@
 # Licensed under the MIT License.
 
 """
-OpenAI extensions for Microsoft Agent 365 Tooling SDK
+OpenAI extensions for Microsoft Agent 365 Tooling SDK.
 
 Tooling and utilities specifically for OpenAI framework integration.
-Provides OpenAI-specific helper utilities.
+Provides OpenAI-specific helper utilities including:
+- McpToolRegistrationService: Service for MCP tool registration and chat history management
+
+For type hints, use the types directly from the OpenAI Agents SDK:
+- agents.memory.Session: Protocol for session objects
+- agents.items.TResponseInputItem: Type for input message items
 """
 
+from .mcp_tool_registration_service import McpToolRegistrationService
+
 __version__ = "1.0.0"
+
+__all__ = [
+    "McpToolRegistrationService",
+]