Serena

Instructions

Implementation Reference

• src/serena/tools/symbol_tools.py:80-149 (handler)

  The FindSymbolTool class provides the core handler implementation for the 'find_symbol' MCP tool. The 'apply' method executes the tool logic by querying the language server symbol retriever for matching symbols, sanitizing the results, converting to JSON, and applying length limits.
  class FindSymbolTool(Tool, ToolMarkerSymbolicRead): """ Performs a global (or local) search using the language server backend. """ # noinspection PyDefaultArgument def apply( self, name_path_pattern: str, depth: int = 0, relative_path: str = "", include_body: bool = False, include_kinds: list[int] = [], # noqa: B006 exclude_kinds: list[int] = [], # noqa: B006 substring_matching: bool = False, max_answer_chars: int = -1, ) -> str: """ Retrieves information on all symbols/code entities (classes, methods, etc.) based on the given name path pattern. The returned symbol information can be used for edits or further queries. Specify `depth > 0` to also retrieve children/descendants (e.g., methods of a class). A name path is a path in the symbol tree *within a source file*. For example, the method `my_method` defined in class `MyClass` would have the name path `MyClass/my_method`. If a symbol is overloaded (e.g., in Java), a 0-based index is appended (e.g. "MyClass/my_method[0]") to uniquely identify it. To search for a symbol, you provide a name path pattern that is used to match against name paths. It can be * a simple name (e.g. "method"), which will match any symbol with that name * a relative path like "class/method", which will match any symbol with that name path suffix * an absolute name path "/class/method" (absolute name path), which requires an exact match of the full name path within the source file. Append an index `[i]` to match a specific overload only, e.g. "MyClass/my_method[1]". :param name_path_pattern: the name path matching pattern (see above) :param depth: depth up to which descendants shall be retrieved (e.g. use 1 to also retrieve immediate children; for the case where the symbol is a class, this will return its methods). Default 0. :param relative_path: Optional. Restrict search to this file or directory. If None, searches entire codebase. If a directory is passed, the search will be restricted to the files in that directory. If a file is passed, the search will be restricted to that file. If you have some knowledge about the codebase, you should use this parameter, as it will significantly speed up the search as well as reduce the number of results. :param include_body: If True, include the symbol's source code. Use judiciously. :param include_kinds: Optional. List of LSP symbol kind integers to include. (e.g., 5 for Class, 12 for Function). Valid kinds: 1=file, 2=module, 3=namespace, 4=package, 5=class, 6=method, 7=property, 8=field, 9=constructor, 10=enum, 11=interface, 12=function, 13=variable, 14=constant, 15=string, 16=number, 17=boolean, 18=array, 19=object, 20=key, 21=null, 22=enum member, 23=struct, 24=event, 25=operator, 26=type parameter. If not provided, all kinds are included. :param exclude_kinds: Optional. List of LSP symbol kind integers to exclude. Takes precedence over `include_kinds`. If not provided, no kinds are excluded. :param substring_matching: If True, use substring matching for the last element of the pattern, such that "Foo/get" would match "Foo/getValue" and "Foo/getData". :param max_answer_chars: Max characters for the JSON result. If exceeded, no content is returned. -1 means the default value from the config will be used. :return: a list of symbols (with locations) matching the name. """ parsed_include_kinds: Sequence[SymbolKind] | None = [SymbolKind(k) for k in include_kinds] if include_kinds else None parsed_exclude_kinds: Sequence[SymbolKind] | None = [SymbolKind(k) for k in exclude_kinds] if exclude_kinds else None symbol_retriever = self.create_language_server_symbol_retriever() symbols = symbol_retriever.find( name_path_pattern, include_kinds=parsed_include_kinds, exclude_kinds=parsed_exclude_kinds, substring_matching=substring_matching, within_relative_path=relative_path, ) symbol_dicts = [_sanitize_symbol_dict(s.to_dict(kind=True, location=True, depth=depth, include_body=include_body)) for s in symbols] result = self._to_json(symbol_dicts) return self._limit_length(result, max_answer_chars)
• src/serena/tools/tools_base.py:119-125 (registration)

  The get_name_from_cls method derives the MCP tool name 'find_symbol' from the class name 'FindSymbolTool' by removing 'Tool' suffix and converting to snake_case.
  def get_name_from_cls(cls) -> str: name = cls.__name__ if name.endswith("Tool"): name = name[:-4] # convert to snake_case name = "".join(["_" + c.lower() if c.isupper() else c for c in name]).lstrip("_") return name
• src/serena/tools/tools_base.py:355-366 (registration)

  The ToolRegistry singleton automatically discovers all Tool subclasses in serena.tools packages, computes their names using get_name_from_cls, and registers 'find_symbol' along with all other tools.
  @singleton class ToolRegistry: def __init__(self) -> None: self._tool_dict: dict[str, RegisteredTool] = {} for cls in iter_subclasses(Tool): if not any(cls.__module__.startswith(pkg) for pkg in tool_packages): continue is_optional = issubclass(cls, ToolMarkerOptional) name = cls.get_name_from_cls() if name in self._tool_dict: raise ValueError(f"Duplicate tool name found: {name}. Tool classes must have unique names.") self._tool_dict[name] = RegisteredTool(tool_class=cls, is_optional=is_optional, tool_name=name)
• src/serena/tools/symbol_tools.py:20-33 (helper)

  Helper function used by FindSymbolTool.apply to sanitize symbol dictionaries before returning as JSON.
  def _sanitize_symbol_dict(symbol_dict: dict[str, Any]) -> dict[str, Any]: """ Sanitize a symbol dictionary inplace by removing unnecessary information. """ # We replace the location entry, which repeats line information already included in body_location # and has unnecessary information on column, by just the relative path. symbol_dict = copy(symbol_dict) s_relative_path = symbol_dict.get("location", {}).get("relative_path") if s_relative_path is not None: symbol_dict["relative_path"] = s_relative_path symbol_dict.pop("location", None) # also remove name, name_path should be enough symbol_dict.pop("name") return symbol_dict
• src/serena/mcp.py:252-259 (registration)

  During MCP server startup, _set_mcp_tools iterates over exposed tool instances from the agent, creates MCPTool instances using make_mcp_tool (which uses tool.get_name() as the tool name), and registers them in the MCP tool manager by name, including 'find_symbol'.
  def _set_mcp_tools(self, mcp: FastMCP, openai_tool_compatible: bool = False) -> None: """Update the tools in the MCP server""" if mcp is not None: mcp._tool_manager._tools = {} for tool in self._iter_tools(): mcp_tool = self.make_mcp_tool(tool, openai_tool_compatible=openai_tool_compatible) mcp._tool_manager._tools[tool.get_name()] = mcp_tool log.info(f"Starting MCP server with {len(mcp._tool_manager._tools)} tools: {list(mcp._tool_manager._tools.keys())}")