more fixes

Author: Benny Chen

eval_protocol/mcp/client/connection.py

@@ -11,7 +11,7 @@
 import logging
 import time
 from contextlib import AsyncExitStack
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any, Dict, List, Optional, Tuple, cast
 
 import httpx
 from mcp.client.session import ClientSession
@@ -276,7 +276,10 @@ async def _get_initial_state_from_mcp_resource(self, session: MCPSession) -> Any
         try:
             # List available resources - this is where initial state should come from
             logger.debug(f"Session {session.session_id}: Discovering MCP resources for initial state...")
-            resources_response = await mcp_session.list_resources()
+            mcp_session_local = session._mcp_session
+            if mcp_session_local is None:
+                raise RuntimeError("Session not initialized while listing resources")
+            resources_response = await mcp_session_local.list_resources()
             resources = resources_response.resources if hasattr(resources_response, "resources") else []
             logger.debug(f"Session {session.session_id}: Found {len(resources)} MCP resources")
             for resource in resources:
@@ -303,7 +306,10 @@ async def _get_initial_state_from_mcp_resource(self, session: MCPSession) -> Any
                     f"Session {session.session_id}: Reading initial state from resource: {initial_state_resource.uri}"
                 )
 
-                resource_content = await mcp_session.read_resource(initial_state_resource.uri)
+                mcp_session_for_read = session._mcp_session
+                if mcp_session_for_read is None:
+                    raise RuntimeError("Session not initialized while reading resource")
+                resource_content = await mcp_session_for_read.read_resource(initial_state_resource.uri)
 
                 # Handle the new ResourceContents format
                 text_value = getattr(resource_content, "text", None)
@@ -348,7 +354,10 @@ async def _get_initial_state_from_mcp_resource(self, session: MCPSession) -> Any
                         f"Session {session.session_id}: About to call mcp_session.read_resource with fallback URI: {first_resource.uri}"
                     )
 
-                    resource_content = await mcp_session.read_resource(first_resource.uri)
+                    mcp_session_for_fallback_read = session._mcp_session
+                    if mcp_session_for_fallback_read is None:
+                        raise RuntimeError("Session not initialized while reading fallback resource")
+                    resource_content = await mcp_session_for_fallback_read.read_resource(first_resource.uri)
 
                     logger.debug(
                         f"Session {session.session_id}: fallback read_resource returned type: {type(resource_content)}"

eval_protocol/mcp_agent/intermediary_server.py

@@ -136,7 +136,11 @@ def make(
 
     if evaluation_rows:
         for i, row in enumerate(evaluation_rows):
-            dataset_info = row.input_metadata.dataset_info if row.input_metadata else {}
+            dataset_info = (
+                row.input_metadata.dataset_info
+                if (row.input_metadata and row.input_metadata.dataset_info is not None)
+                else {}
+            )
 
             system_message = row.get_system_message()
             system_prompt = system_message.content or ""

eval_protocol/mcp_agent/orchestration/remote_http_client.py

@@ -68,6 +68,7 @@ def close(self):
 
     def _execute_airline_action(self, action_name: str, parameters: Dict[str, Any]) -> Dict[str, Any]:
         """Execute action using airline tools."""
+        assert isinstance(self.airline_tools, AirlineTools), "Airline tools not initialized"
         action_map = {
             "book_reservation": self.airline_tools.book_reservation,
             "cancel_reservation": self.airline_tools.cancel_reservation,

eval_protocol/mcp_env.py

@@ -224,7 +224,7 @@ async def __call__(
         tool_schemas: List[Dict],
         env_index: int,
         conversation_history: List[Dict[str, Any]],
-    ):
+    ) -> Tuple[List["MCPToolCall"], Optional[Dict[str, int]], Optional[str]]:
         """
         Main policy call method. Delegates to playback or live mode.
 

eval_protocol/mcp_servers/tau2/airplane_environment/airline_environment.py

@@ -608,7 +608,7 @@ def _ioi_cpp_code_reward_impl(
             },
         )
 
-    response_content = messages[-1].content
+    response_content = messages[-1].content if isinstance(messages[-1].content, str) else ""
 
     expected_output_str_from_gt: Optional[str] = None
     test_cases_from_gt: Optional[List[Dict[str, Any]]] = None