fix tests

Author: Benny Chen

eval_protocol/benchmarks/test_livebench_data_analysis.py

@@ -287,6 +287,8 @@ def _read_jsonl_table_from_text(text: str, header_cols: List[str]):
             return 0
 
     # Compare
+    assert llm_df is not None, "LLM dataframe is None"
+    assert gt_df is not None, "GT dataframe is None"
     try:
         gt_df.columns = [str(s).strip() for s in gt_df.columns]
         if "index" in gt_df.columns:

eval_protocol/mcp/mcpgym.py

@@ -563,7 +563,7 @@ def format_observation(self, obs: Any, env: Any) -> Dict[str, Any]:
         else:
             return {"observation": serialized_obs}
 
-    def run(self, transport: str = "streamable-http", **kwargs):
+    def run(self, transport: Literal["stdio", "sse", "streamable-http"] = "streamable-http", **kwargs):
         """Run the unified MCP-Gym server with high concurrency settings."""
         if transport == "streamable-http":
             # Run with custom high-concurrency uvicorn config

eval_protocol/mcp/simulation_server.py

@@ -30,7 +30,8 @@ def reset_environment(self, env, seed): ...
 from abc import ABC, abstractmethod
 from collections.abc import AsyncIterator
 from contextlib import asynccontextmanager
-from typing import Any, Callable, Dict, List, Optional, Tuple
+from typing import Any, Awaitable, Callable, Dict, List, Optional, Tuple, Iterable
+from pydantic import AnyUrl
 
 import uvicorn
 from mcp.server.lowlevel import Server

eval_protocol/mcp_agent/orchestration/local_docker_client.py

@@ -650,7 +650,7 @@ async def list_tools_on_instance(self, instance: ManagedInstanceInfo) -> types.L
                 )
             target_base_url = instance.mcp_endpoint_url.rstrip("/")
             try:
-                async with streamablehttp_client(url=target_base_url) as (
+                async with streamablehttp_client(base_url=target_base_url) as (  # type: ignore
                     read_s,
                     write_s,
                     _,  # get_session_id_func usually not needed for a single call