fix: restore CI to green (pin mcp<2, unmask and fix a racy test)

pyproject.toml

@@ -22,14 +22,14 @@ dev = [
   "httpx>=0.27",
   "fastapi>=0.110",
   "uvicorn>=0.29",
-  "mcp>=1.2",
+  "mcp>=1.2,<2",
 ]
 api = [
   "fastapi>=0.110",
   "uvicorn>=0.29",
 ]
 mcp = [
-  "mcp>=1.2",
+  "mcp>=1.2,<2",
 ]
 render = [
   "manim>=0.19",

tests/test_backends.py

@@ -35,6 +35,13 @@ class ResponseBody:
         def read(self):
             return b'{"error":"bad auth"}'
 
+        def close(self):
+            # HTTPError wraps fp in a finalizer that calls close() on GC. Without
+            # this, the AttributeError surfaces as an unraisable exception inside
+            # whatever unrelated test happens to be running when the collector
+            # gets to it.
+            pass
+
     def raise_httperror(*args, **kwargs):
         raise HTTPError(
             url="https://example.test/v1/chat/completions",

tests/test_sol_staged_pipeline.py

@@ -3,6 +3,7 @@
 import subprocess
 import sys
 import threading
+import time
 from pathlib import Path
 
 import pytest
@@ -66,9 +67,21 @@ def __init__(self):
             self.returncode = 0
 
         def wait(self, timeout=None):
-            observed["trace_during_wait"] = (
-                tmp_path / "trace.jsonl"
-            ).read_text(encoding="utf-8")
+            # CodexCli streams stdout to the trace from a daemon thread it only
+            # joins after wait() returns, so nothing orders that thread against
+            # this call. Poll instead of reading once: the point of the
+            # assertion is that the trace fills during the run rather than
+            # afterwards, and a bare read just races the reader thread.
+            trace = tmp_path / "trace.jsonl"
+            deadline = time.monotonic() + 10.0
+            text = ""
+            while time.monotonic() < deadline:
+                if trace.is_file():
+                    text = trace.read_text(encoding="utf-8")
+                    if "thread.started" in text:
+                        break
+                time.sleep(0.01)
+            observed["trace_during_wait"] = text
             return self.returncode
 
         def kill(self):