fix(tools): keep MCP tool names unique for non-ASCII inputs

nanobot/agent/tools/mcp.py

@@ -173,23 +173,37 @@ def _filter_malformed_mcp_progress_notifications(read_stream: Any, server_name:
     return _MalformedProgressNotificationFilter(read_stream, server_name)
 
 
-def _sanitize_name(name: str) -> str:
-    """Sanitize an MCP-derived name for model API compatibility."""
-    return _SANITIZE_RE.sub("_", re.sub(r"[^a-zA-Z0-9_-]", "_", name))
-
-
 _MAX_TOOL_NAME_LENGTH = 64
 _HASH_LENGTH = 8
 
 
+def _short_digest(value: str, length: int = _HASH_LENGTH) -> str:
+    """Return a truncated hex digest of ``value`` for naming purposes.
+
+    Not for security; just produces a stable, collision-resistant short suffix
+    when a name must be shortened or replaced.
+    """
+    return hashlib.sha1(value.encode("utf-8")).hexdigest()[:length]
+
+
+def _sanitize_name(name: str) -> str:
+    """Sanitize an MCP-derived name for model API compatibility."""
+    sanitized = _SANITIZE_RE.sub("_", re.sub(r"[^a-zA-Z0-9_-]", "_", name))
+    if re.search(r"[^\x00-\x7f]", name):
+        core = sanitized.rstrip("_")
+        if core == "":
+            return f"tool_{_short_digest(name)}"
+        return f"{core}_{_short_digest(name)}"
+    return sanitized
+
+
 def _limit_tool_name(name: str, max_length: int = _MAX_TOOL_NAME_LENGTH) -> str:
     """Limit a tool name while keeping short names unchanged."""
     if len(name) <= max_length:
         return name
 
-    digest = hashlib.sha1(name.encode("utf-8")).hexdigest()[:_HASH_LENGTH]
     prefix_length = max_length - _HASH_LENGTH - 1
-    return f"{name[:prefix_length]}_{digest}"
+    return f"{name[:prefix_length]}_{_short_digest(name)}"
 
 
 def _sanitize_mcp_tool_name(name: str) -> str:

tests/tools/test_mcp_tool.py

@@ -1812,6 +1812,48 @@ def test_sanitize_name_noop_for_already_clean_names() -> None:
     assert _sanitize_name("mcp_server_tool") == "mcp_server_tool"
 
 
+def test_sanitize_name_falls_back_to_hash_for_all_underscore_names() -> None:
+    # Fully non-ASCII names collapse to underscores; they must fall back to a
+    # deterministic hash-based name instead of silently colliding on "_".
+    import hashlib
+
+    name = "获取天气"
+    expected = f"tool_{hashlib.sha1(name.encode('utf-8')).hexdigest()[:8]}"
+    assert _sanitize_name(name) == expected
+
+
+def test_sanitize_name_keeps_distinct_non_ascii_names_unique() -> None:
+    # Two different non-ASCII names must sanitize to different results so they
+    # don't collide when registered.
+    assert _sanitize_name("获取天气") != _sanitize_name("日本語ツール")
+
+
+def test_sanitize_name_hashes_non_ascii_with_ascii_prefix() -> None:
+    # The real call path always carries an ASCII ``mcp_{server}_`` prefix, so
+    # the old ``tool_{digest}`` fallback never fired and Chinese tool names
+    # collapsed to the same underscore run. The digest must be appended to the
+    # stripped core instead, without stacking underscores.
+    name = "mcp_server_城市天气实况"
+    expected = f"mcp_server_{mcp_mod._short_digest(name)}"
+    assert _sanitize_name(name) == expected
+
+
+def test_sanitize_name_keeps_ascii_part_for_mixed_names() -> None:
+    # Mixed Chinese+ASCII names keep their readable ASCII part (15, POI, ...)
+    # in front of the digest.
+    name = "mcp_天气预报_城市15日预报"
+    assert _sanitize_name(name) == f"mcp_15_{mcp_mod._short_digest(name)}"
+    assert "__" not in _sanitize_name(name)
+
+
+def test_sanitize_name_leaves_pure_ascii_names_untouched() -> None:
+    # Only non-ASCII names get a digest appended; ASCII names with spaces or
+    # special characters sanitize readably without a hash.
+    assert _sanitize_name("mcp_srv_My Tool") == "mcp_srv_My_Tool"
+    assert _sanitize_name("mcp_my server_prompt_design-schema") == "mcp_my_server_prompt_design-schema"
+    assert _sanitize_name("mcp_server_tool") == "mcp_server_tool"
+
+
 # ---------------------------------------------------------------------------
 # Wrapper sanitization tests
 # ---------------------------------------------------------------------------