From 16322651bb8a18c7a4d43a539dac308bbed02c50 Mon Sep 17 00:00:00 2001
From: L4XB <lukas.buck@e-mail.de>
Date: Thu, 10 Sep 2026 22:45:09 +0200
Subject: [PATCH] fix: do not load the anthropic package on import

`serena.analytics` imported `anthropic.types` at module level for two type
annotations, so every `import serena.cli` / `import serena.agent` paid for
loading the whole anthropic package (2-4 s on the reporter's Windows machine)
even with the default CHAR_COUNT token estimator. That delay can push MCP
startup past a client's initial tool-discovery window.

The import now lives behind `TYPE_CHECKING`; `AnthropicTokenCount` already
imported `anthropic` lazily in `__init__`, and the request body is a plain
dict (what `MessageParam` produces at runtime anyway).

Tests: importing serena.cli / serena.agent / serena.analytics in a fresh
interpreter leaves `anthropic` out of `sys.modules`; the Anthropic estimator
still sends the expected count_tokens request.

Fixes #2012
---
 CHANGELOG.md                          |  2 ++
 src/serena/analytics.py               |  9 +++++++--
 test/serena/test_analytics_imports.py | 29 +++++++++++++++++++++++++++
 3 files changed, 38 insertions(+), 2 deletions(-)
 create mode 100644 test/serena/test_analytics_imports.py

diff --git a/CHANGELOG.md b/CHANGELOG.md
index a6aa0625..56e3b8c4 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -13,6 +13,8 @@ Status of the `main` branch. Changes prior to the next official version change w
 
 * General:
   - Fix: MCP `initialize` now reports Serena's version instead of the installed mcp SDK version (#1889)
+  - Fix: importing Serena no longer loads the `anthropic` package unless the Anthropic token counter is
+    actually used; the unconditional import added seconds to CLI/MCP startup on some machines (#2012)
   - Fix: Parallel agents auto-registering projects could overwrite each other's changes to the global
     project list in `serena_config.yml`
   - Fix: `TextUtils.insert_text_at_position` returned a wrong position when the inserted text merged
diff --git a/src/serena/analytics.py b/src/serena/analytics.py
index 315cd57d..8fcc5636 100644
--- a/src/serena/analytics.py
+++ b/src/serena/analytics.py
@@ -9,10 +9,15 @@ from collections import defaultdict
 from copy import copy
 from dataclasses import asdict, dataclass
 from enum import Enum
+from typing import TYPE_CHECKING
 
-from anthropic.types import MessageParam, MessageTokensCount
 from dotenv import load_dotenv
 
+if TYPE_CHECKING:
+    # Imported for annotations only: loading the anthropic package costs seconds on some
+    # machines (see #2012) and is only needed when the Anthropic token counter is used.
+    from anthropic.types import MessageTokensCount
+
 log = logging.getLogger(__name__)
 
 
@@ -64,7 +69,7 @@ class AnthropicTokenCount(TokenCountEstimator):
     def _send_count_tokens_request(self, text: str) -> MessageTokensCount:
         return self._anthropic_client.messages.count_tokens(
             model=self._model_name,
-            messages=[MessageParam(role="user", content=text)],
+            messages=[{"role": "user", "content": text}],
         )
 
     def estimate_token_count(self, text: str) -> int:
diff --git a/test/serena/test_analytics_imports.py b/test/serena/test_analytics_imports.py
new file mode 100644
index 00000000..ec18f378
--- /dev/null
+++ b/test/serena/test_analytics_imports.py
@@ -0,0 +1,29 @@
+"""The anthropic package must only be loaded when the Anthropic token counter is used (#2012)."""
+
+import subprocess
+import sys
+from unittest.mock import MagicMock
+
+import pytest
+
+from serena.analytics import AnthropicTokenCount
+
+
+@pytest.mark.parametrize("module", ["serena.cli", "serena.agent", "serena.analytics"])
+def test_importing_serena_does_not_load_anthropic(module: str) -> None:
+    code = f"import sys, {module}; print('anthropic' in sys.modules)"
+    result = subprocess.run([sys.executable, "-c", code], capture_output=True, text=True, check=True, timeout=120)
+    assert result.stdout.strip() == "False", f"importing {module} loaded the anthropic package"
+
+
+def test_anthropic_token_count_sends_a_plain_user_message() -> None:
+    estimator = AnthropicTokenCount.__new__(AnthropicTokenCount)
+    estimator._model_name = "claude-sonnet-4-20250514"
+    estimator._anthropic_client = MagicMock()
+    estimator._anthropic_client.messages.count_tokens.return_value = MagicMock(input_tokens=7)
+
+    assert estimator.estimate_token_count("hello") == 7
+    estimator._anthropic_client.messages.count_tokens.assert_called_once_with(
+        model="claude-sonnet-4-20250514",
+        messages=[{"role": "user", "content": "hello"}],
+    )