fix more type errors

Author: Benny Chen

eval_protocol/mcp/client/connection.py

@@ -53,16 +53,26 @@ async def initialize_session(self, session: MCPSession) -> None:
 
         exit_stack = AsyncExitStack()
 
-        client_info = Implementation(name="reward-kit", version="1.0.0", _extra={})
-        client_info._extra["session_id"] = session.session_id
+        # Attach client metadata for the server to consume (session_id, seed, config, etc.).
+        # The server inspects a private `_extra` dict on client_info, so we populate it here.
+        client_info = Implementation(name="reward-kit", version="1.0.0")
+        extra_data: Dict[str, Any] = {"session_id": session.session_id}
         if session.seed is not None:
-            client_info._extra["seed"] = session.seed
+            extra_data["seed"] = session.seed
         if session.dataset_row and session.dataset_row.environment_context:
-            client_info._extra["config"] = session.dataset_row.environment_context
+            extra_data["config"] = session.dataset_row.environment_context
         if session.dataset_row and session.dataset_row.id:
-            client_info._extra["dataset_row_id"] = session.dataset_row.id
+            extra_data["dataset_row_id"] = session.dataset_row.id
         if session.model_id:
-            client_info._extra["model_id"] = session.model_id
+            extra_data["model_id"] = session.model_id
+
+        # Merge with any existing _extra dict instead of overwriting
+        existing_extra = getattr(client_info, "_extra", None)
+        merged_extra: Dict[str, Any] = {}
+        if isinstance(existing_extra, dict):
+            merged_extra.update(existing_extra)
+        merged_extra.update(extra_data)
+        setattr(client_info, "_extra", merged_extra)
 
         read_stream, write_stream, _ = await exit_stack.enter_async_context(
             streamablehttp_client(session.base_url, terminate_on_close=True)
@@ -92,7 +102,10 @@ async def _prewarm_tools_cache(self, session: MCPSession) -> None:
             # Only fetch tools if not already cached for this base_url
             if cache_key not in self._tools_cache:
                 logger.debug(f"Pre-warming tools cache for {cache_key}")
-                tools_response = await session._mcp_session.list_tools()
+                mcp_session_local = session._mcp_session
+                if mcp_session_local is None:
+                    raise RuntimeError("Session not initialized during prewarm")
+                tools_response = await mcp_session_local.list_tools()
                 tools = tools_response.tools if hasattr(tools_response, "tools") else []
 
                 tool_schemas = []
@@ -213,7 +226,7 @@ async def get_initial_state(self, session: MCPSession) -> Any:
                 try:
                     # Use shorter timeout for playback mode, longer timeout for high-concurrency initialization
                     # (50+ concurrent sessions need more time for initial state setup)
-                    timeout = 3.0 if hasattr(session, "_is_playback_mode") and session._is_playback_mode else 15.0
+                    timeout = 3.0 if bool(getattr(session, "_is_playback_mode", False)) else 15.0
                     async with httpx.AsyncClient(timeout=timeout) as client:
                         initial_state_response = await client.get(
                             f"{base_url}/control/initial_state",

eval_protocol/mcp/clients.py

@@ -29,7 +29,7 @@ def __init__(self, intermediary_server_url: str):
 
     async def connect(self):
         """Establishes connection and MCP session."""
-        if self._mcp_session and not self._mcp_session.is_closed:
+        if self._mcp_session is not None:
             logger.debug("Already connected.")
             return
 

eval_protocol/mcp/mcp_multi_client.py

@@ -90,8 +90,8 @@ async def _connect_to_server(
             if env_config:
                 self._validate_environment_variables(server_name, env_config)
 
-            # Use the current system environment (os.environ) - don't override with config
-            server_params = StdioServerParameters(command=command, args=args, env=os.environ)
+            # Use the current system environment (os.environ) - convert to plain dict for typing compatibility
+            server_params = StdioServerParameters(command=command, args=args, env=dict(os.environ))
 
             stdio_transport = await self.exit_stack.enter_async_context(stdio_client(server_params))
             stdio, write = stdio_transport

eval_protocol/mcp_agent/orchestration/local_docker_client.py

@@ -244,6 +244,10 @@ async def provision_instances(
                 logger.info(
                     f"Creating template container for commit: {temp_cont_name} from {backend_config.docker_image}"
                 )
+                if not backend_config.docker_image:
+                    raise ValueError(
+                        f"docker_image is required for template commit for backend {backend_config.backend_name_ref}"
+                    )
                 temp_c = self.docker_client.containers.run(  # type: ignore
                     image=backend_config.docker_image,
                     name=temp_cont_name,
@@ -322,6 +326,11 @@ async def provision_instances(
                 logger.info(
                     f"Provisioning instance {container_name} (transport: {backend_config.mcp_transport}) from image {image_to_run_from}"
                 )
+                # Ensure the image reference is present before using it in Docker APIs
+                if not image_to_run_from:
+                    raise ValueError(
+                        f"docker_image is required to provision instance {container_name} for backend {backend_config.backend_name_ref}"
+                    )
                 if backend_config.mcp_transport == "http":
                     # ... (HTTP provisioning logic, ensure it uses current_container_volumes) ...
                     if not self.docker_client:
@@ -640,7 +649,7 @@ async def list_tools_on_instance(self, instance: ManagedInstanceInfo) -> types.L
                 )
             target_base_url = instance.mcp_endpoint_url.rstrip("/")
             try:
-                async with streamablehttp_client(base_url=target_base_url) as (
+                async with streamablehttp_client(target_base_url) as (
                     read_s,
                     write_s,
                     _,  # get_session_id_func usually not needed for a single call

eval_protocol/mcp_agent/orchestration/remote_http_client.py

@@ -263,6 +263,9 @@ async def call_tool_on_instance(
             logger.debug(f"Proxying tool {tool_name} to {target_url} for instance {instance.instance_id}")
         else:
             # Call tool directly on the instance's MCP endpoint
+            # mypy/pyright: instance.mcp_endpoint_url is Optional[str]; validate before assignment
+            if not instance.mcp_endpoint_url:
+                raise ValueError(f"Instance {instance.instance_id} missing mcp_endpoint_url for direct tool call")
             target_url = instance.mcp_endpoint_url
             logger.debug(f"Calling tool {tool_name} directly on {target_url} for instance {instance.instance_id}")
 

eval_protocol/models.py

@@ -611,27 +611,35 @@ def get_steps(self) -> int:
     def get_total_reward(self) -> float:
         """Get total reward from control_plane_step data."""
         messages_with_control_plane = [msg for msg in self.messages if msg.control_plane_step]
-        return (
-            sum(msg.control_plane_step["reward"] for msg in messages_with_control_plane)
-            if messages_with_control_plane
-            else 0.0
-        )
+        if not messages_with_control_plane:
+            return 0.0
+        total = 0.0
+        for msg in messages_with_control_plane:
+            step = msg.control_plane_step or {}
+            try:
+                total += float(step.get("reward", 0.0))
+            except (TypeError, ValueError):
+                continue
+        return total
 
     def get_terminated(self) -> bool:
         """Get termination status from control_plane_step data."""
         messages_with_control_plane = [msg for msg in self.messages if msg.control_plane_step]
-        return (
-            any(msg.control_plane_step["terminated"] for msg in messages_with_control_plane)
-            if messages_with_control_plane
-            else False
-        )
+        if not messages_with_control_plane:
+            return False
+        for msg in messages_with_control_plane:
+            step = msg.control_plane_step or {}
+            if bool(step.get("terminated", False)):
+                return True
+        return False
 
     def get_termination_reason(self) -> str:
         """Get termination reason from the final control_plane_step data."""
         # Find the last message with control_plane_step that has termination_reason
         for msg in reversed(self.messages):
             if msg.control_plane_step and msg.control_plane_step.get("termination_reason"):
-                return msg.control_plane_step["termination_reason"]
+                reason = msg.control_plane_step.get("termination_reason")
+                return str(reason)
         return "unknown"
 
     def __hash__(self) -> int:

eval_protocol/rewards/code_execution.py

@@ -171,7 +171,11 @@ def local_code_execution_reward(
             },
         )
 
-    response_content = messages[-1].content
+    # Normalize content to string; Message.content may be str or list of content parts
+    last_content = messages[-1].content
+    response_content = (
+        last_content if isinstance(last_content, str) else "".join([p.text for p in (last_content or [])])
+    )
     expected_output_str = ground_truth
 
     code_blocks = extract_code_blocks(response_content, language)
@@ -935,7 +939,10 @@ def e2b_code_execution_reward(
             },
         )
 
-    response_content = messages[-1].content
+    last_content = messages[-1].content
+    response_content = (
+        last_content if isinstance(last_content, str) else "".join([p.text for p in (last_content or [])])
+    )
     expected_output_str = ground_truth
 
     code_blocks = extract_code_blocks(response_content, language)

eval_protocol/rewards/lean_prover.py

@@ -57,7 +57,8 @@ def lean_prover_reward(
             },
         )
 
-    response = messages[-1].content
+    last_content = messages[-1].content
+    response = last_content if isinstance(last_content, str) else "".join([p.text for p in (last_content or [])])
     if not response:
         return EvaluateResult(
             score=0.0,
@@ -230,7 +231,10 @@ def deepseek_prover_v2_reward(
         and messages[-1].role == "assistant"
         and messages[-1].content is not None
     ):
-        response_content = messages[-1].content
+        last_content = messages[-1].content
+        response_content = (
+            last_content if isinstance(last_content, str) else "".join([p.text for p in (last_content or [])])
+        )
 
     final_score = base_score
     subgoal_count = 0