satif-ai 0.2.9__py3-none-any.whl → 0.2.11__py3-none-any.whl

satif_ai/adapters/tidy.py

@@ -9,7 +9,7 @@ from pathlib import Path
 from typing import Optional, Union
 
 from agents import Agent, Runner, function_tool
-from agents.mcp.server import MCPServerStdio
+from agents.mcp.server import MCPServer
 from mcp import ClientSession
 from satif_core.adapters.base import Adapter
 from satif_core.types import Datasource, SDIFPath
@@ -224,7 +224,7 @@ class TidyAdapter(Adapter):
 
     def __init__(
         self,
-        mcp_server: MCPServerStdio,
+        mcp_server: MCPServer,
         mcp_session: ClientSession,
         llm_model: str = "o4-mini",
         max_iterations: int = 5,
@@ -233,7 +233,7 @@ class TidyAdapter(Adapter):
         Initialize the TidyAdapter.
 
         Args:
-            mcp_server: An instance of MCPServerStdio for agent communication.
+            mcp_server: An instance of MCPServer for agent communication.
             mcp_session: An instance of ClientSession for resource/prompt fetching.
             llm_model: Name of the language model to use for the agent.
             max_iterations: Maximum number of attempts the agent gets to refine the code.
@@ -349,7 +349,7 @@ class TidyAdapter(Adapter):
         if isinstance(sdif, SDIFDatabase):
             input_path = Path(sdif.path)
         else:
-            input_path = sdif
+            input_path = Path(sdif)
         if not input_path.exists():
             raise FileNotFoundError(f"Input SDIF file not found: {input_path}")
 

satif_ai/standardizers/ai.py

@@ -11,6 +11,7 @@ from satif_core.standardizers.base import AsyncStandardizer
 from satif_core.types import Datasource, FilePath, SDIFPath, StandardizationResult
 
 from satif_ai.adapters.tidy import TidyAdapter
+from satif_ai.standardizers.ai_xlsx import AIXLSXStandardizer
 from satif_ai.utils.merge_sdif import merge_sdif_files
 from satif_ai.utils.zip import extract_zip_archive_async
 
@@ -43,8 +44,9 @@ class AIStandardizer(AsyncStandardizer):
 
         self.ai_standardizer_map: Dict[str, Type[AsyncStandardizer]] = {
             ".csv": AICSVStandardizer,
-            # Future standardizers:
-            # ".xlsx": AIXLSXStandardizer,
+            ".xlsx": AIXLSXStandardizer,
+            ".xls": AIXLSXStandardizer,
+            ".xlsm": AIXLSXStandardizer,
             # ".pdf": AIPDFStandardizer,
             # ".json": AIJSONStandardizer,
             # ".xml": AIXMLStandardizer,
@@ -332,11 +334,9 @@ class AIStandardizer(AsyncStandardizer):
             logger.info(
                 f"Merging {len(intermediate_sdif_files)} intermediate SDIF SQLite files into '{final_sdif_file_target}'."
             )
-            # merge_sdif_files must accept a list of source SQLite file paths and a target SQLite file path.
-            merged_target_path = await merge_sdif_files(
+            merged_target_path = merge_sdif_files(
                 intermediate_sdif_files,
                 final_sdif_file_target,
-                overwrite=False,  # We handled overwrite for final_sdif_file_target
             )
             final_sdif_path_str = str(merged_target_path)
 
@@ -414,13 +414,11 @@ class AIStandardizer(AsyncStandardizer):
         file_processing_temp_dir.mkdir(parents=True, exist_ok=True)
 
         try:
-            # 1. Resolve input datasource to a list of processable file paths
             resolved_files = await self._resolve_input_files(
                 datasource, file_processing_temp_dir
             )
             logger.info(f"Resolved {len(resolved_files)} file(s) for standardization.")
 
-            # 2. Group files by the AI standardizer responsible for them
             grouped_by_std, unsupported = self._group_files_by_standardizer(
                 resolved_files
             )
@@ -438,7 +436,6 @@ class AIStandardizer(AsyncStandardizer):
                 f"File groups for standardization: { {cls.__name__: [f.name for f in paths] for cls, paths in grouped_by_std.items()} }"
             )
 
-            # 3. Process each group of files, generating intermediate SDIF SQLite files
             (
                 intermediate_sdif_files,
                 aggregated_file_configs,
@@ -454,7 +451,6 @@ class AIStandardizer(AsyncStandardizer):
                 f"Successfully generated {len(intermediate_sdif_files)} intermediate SDIF SQLite file(s)."
             )
 
-            # 4. Consolidate intermediate SDIF files into the final target file
             final_result = await self._consolidate_results(
                 intermediate_sdif_files,
                 aggregated_file_configs,

satif_ai/standardizers/ai_csv.py

@@ -12,6 +12,7 @@ from agents import Agent, Runner, function_tool
 from agents.mcp.server import MCPServerStdio
 from charset_normalizer import detect
 from mcp import ClientSession
+from satif_core import AsyncStandardizer
 from satif_core.types import Datasource, SDIFPath, StandardizationResult
 from satif_sdk.standardizers.csv import (
     CSVStandardizer,
@@ -274,7 +275,9 @@ async def read_raw_lines(
 
 
 # --- AICSVStandardizer Class ---
-class AICSVStandardizer(CSVStandardizer):  # Inherits from the enhanced CSVStandardizer
+class AICSVStandardizer(
+    CSVStandardizer, AsyncStandardizer
+):  # Inherits from the enhanced CSVStandardizer
     def __init__(
         self,
         mcp_server: Optional[MCPServerStdio] = None,

satif_ai/standardizers/ai_xlsx.py

@@ -0,0 +1,372 @@
+import logging
+import shutil
+import tempfile
+import uuid
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Tuple
+
+try:
+    from xlsx_to_sdif.graph import graph as xlsx_graph
+    from xlsx_to_sdif.state import State as XLSXState
+except ImportError:
+    xlsx_graph = None  # type: ignore
+    XLSXState = None  # type: ignore
+    logging.getLogger(__name__).warning(
+        "Failed to import xlsx_to_sdif. AIXLSXStandardizer will not be functional."
+    )
+
+
+from satif_core.standardizers.base import AsyncStandardizer
+from satif_core.types import Datasource, SDIFPath, StandardizationResult
+
+from satif_ai.utils.merge_sdif import merge_sdif_files
+
+logger = logging.getLogger(__name__)
+
+
+class AIXLSXStandardizer(AsyncStandardizer):
+    """
+    An asynchronous standardizer for XLSX files that leverages the `xlsx-to-sdif` library.
+
+    This standardizer processes one or more XLSX files, converts each to an
+    intermediate SDIF (Standardized Data Interchange Format) file using the
+    `xlsx-to-sdif` processing graph, and then consolidates these intermediate
+    files into a single final SDIF file.
+    """
+
+    def __init__(self, *args: Any, **kwargs: Any):
+        """
+        Initializes the AIXLSXStandardizer.
+
+        Args:
+            ...
+        """
+
+    async def _invoke_xlsx_graph(
+        self, input_file_path: Path, graph_config: Dict[str, Any]
+    ) -> Path:
+        """
+        Invokes the `xlsx-to-sdif` graph for a single XLSX file.
+
+        Args:
+            input_file_path: Path to the input XLSX file.
+            graph_config: Configuration for the `xlsx-to-sdif` graph invocation,
+                          including a unique `thread_id`.
+
+        Returns:
+            Path to the SDIF file produced by the graph.
+
+        Raises:
+            RuntimeError: If the `xlsx-to-sdif` graph is not available, fails to
+                          return a final state, or does not produce an output path.
+            FileNotFoundError: If the graph reports an output file that doesn't exist.
+        """
+        if not xlsx_graph or not XLSXState:
+            raise RuntimeError(
+                "xlsx_to_sdif is not available. "
+                "Please ensure 'xlsx-to-sdif' library is installed correctly."
+            )
+
+        initial_state: XLSXState = {"spreadsheet_path": str(input_file_path)}  # type: ignore
+
+        thread_id = graph_config.get("configurable", {}).get(
+            "thread_id", "unknown_thread"
+        )
+        logger.info(
+            f"Invoking xlsx_to_sdif graph for: {input_file_path.name} with thread_id: {thread_id}"
+        )
+
+        # Stream events for logging or potential progress updates
+        async for event in xlsx_graph.astream_events(
+            initial_state, graph_config, version="v1"
+        ):
+            event_type = event["event"]
+            event_name = event.get("name", "")
+            if event_type in ["on_tool_start", "on_chain_start"]:
+                logger.debug(
+                    f"Graph event for {input_file_path.name} (Thread: {thread_id}): {event_type} - {event_name}"
+                )
+            elif event_type in ["on_tool_error", "on_chain_error", "on_llm_error"]:
+                logger.warning(
+                    f"Graph error event for {input_file_path.name} (Thread: {thread_id}): {event_type} - {event_name}. Data: {event.get('data')}"
+                )
+
+        final_snapshot = await xlsx_graph.aget_state(graph_config)
+        if not final_snapshot or not final_snapshot.values:
+            raise RuntimeError(
+                f"xlsx_to_sdif graph did not return a final state for {input_file_path.name} (Thread: {thread_id})."
+            )
+
+        output_sdif_path_str = final_snapshot.values.get("output_sdif_path")
+        if not output_sdif_path_str:
+            raise RuntimeError(
+                f"xlsx_to_sdif graph for {input_file_path.name} (Thread: {thread_id}) "
+                f"did not produce an 'output_sdif_path' in its final state. State: {final_snapshot.values}"
+            )
+
+        output_sdif_path = Path(output_sdif_path_str)
+        if not output_sdif_path.is_file():
+            raise FileNotFoundError(
+                f"xlsx_to_sdif graph for {input_file_path.name} (Thread: {thread_id}) "
+                f"reported output file '{output_sdif_path}', but it does not exist or is not a file."
+            )
+
+        logger.info(
+            f"xlsx_to_sdif graph successfully processed {input_file_path.name} (Thread: {thread_id}). Output at {output_sdif_path}"
+        )
+        return output_sdif_path
+
+    async def _resolve_and_filter_input_files(
+        self, datasource: Datasource
+    ) -> List[Path]:
+        """Resolves and validates datasource, returning a list of XLSX file paths."""
+        input_files: List[Path]
+        if isinstance(datasource, (str, Path)):
+            input_files = [Path(datasource)]
+        elif isinstance(datasource, list) and all(
+            isinstance(p, (str, Path)) for p in datasource
+        ):
+            input_files = [Path(p) for p in datasource]
+        else:
+            raise ValueError(
+                "Datasource must be a file path (str or Path) or a list of such paths."
+            )
+
+        if not input_files:
+            raise ValueError("No input XLSX files provided in the datasource.")
+
+        xlsx_input_files = []
+        for f_path in input_files:
+            if not f_path.is_file():
+                raise FileNotFoundError(f"Input file not found: {f_path}")
+            if f_path.suffix.lower() not in (
+                ".xlsx",
+                ".xlsm",
+                ".xlsb",
+                ".xls",
+            ):  # Common Excel extensions
+                logger.warning(
+                    f"File {f_path.name} is not a typical XLSX file extension, but will be attempted."
+                )
+            xlsx_input_files.append(f_path)
+
+        if not xlsx_input_files:
+            raise ValueError(
+                "No processable XLSX files found in the datasource after filtering."
+            )
+        return xlsx_input_files
+
+    def _prepare_final_output_path(
+        self, output_path: SDIFPath, overwrite: bool
+    ) -> Path:
+        """Prepares the final output path, handling overwrites and directory creation."""
+        final_output_path = Path(output_path)
+        if final_output_path.exists() and not overwrite:
+            raise FileExistsError(
+                f"Output file {final_output_path} already exists and overwrite is False."
+            )
+        elif final_output_path.exists() and overwrite:
+            logger.info(
+                f"Overwrite active: Deleting existing output file {final_output_path}"
+            )
+            try:
+                if (
+                    final_output_path.is_dir()
+                ):  # Should not happen if SDIFPath is file path
+                    raise IsADirectoryError(
+                        f"Output path {final_output_path} is a directory."
+                    )
+                final_output_path.unlink()
+            except OSError as e:
+                raise RuntimeError(
+                    f"Failed to delete existing output file {final_output_path}: {e}"
+                ) from e
+
+        final_output_path.parent.mkdir(parents=True, exist_ok=True)
+        return final_output_path
+
+    def _setup_temp_directories(self) -> Tuple[Path, Path, Path]:
+        """Creates and returns paths for temporary working directories."""
+        run_temp_dir = Path(tempfile.mkdtemp(prefix="satif_aixlsx_run_"))
+        intermediate_sdif_dir = run_temp_dir / "intermediate_sdifs"
+        intermediate_sdif_dir.mkdir()
+        temp_input_copies_dir = (
+            run_temp_dir / "temp_input_copies"
+        )  # Directory for temporary input copies
+        temp_input_copies_dir.mkdir()
+        return run_temp_dir, intermediate_sdif_dir, temp_input_copies_dir
+
+    async def _process_single_file_to_intermediate_sdif(
+        self,
+        input_xlsx_file: Path,
+        final_output_path_stem: str,
+        temp_input_copies_dir: Path,
+        intermediate_sdif_dir: Path,
+    ) -> Path:
+        """Processes a single XLSX file to an intermediate SDIF in a controlled location."""
+        logger.info(f"Processing file: {input_xlsx_file.name}")
+        graph_thread_id = f"satif_aixlsx_{final_output_path_stem}_{input_xlsx_file.stem}_{uuid.uuid4().hex[:8]}"
+
+        temp_input_file_for_graph = (
+            temp_input_copies_dir
+            / f"{input_xlsx_file.stem}_{graph_thread_id}{input_xlsx_file.suffix}"
+        )
+        shutil.copy2(input_xlsx_file, temp_input_file_for_graph)
+        logger.debug(
+            f"Created temporary copy of {input_xlsx_file.name} at {temp_input_file_for_graph}"
+        )
+
+        graph_config_for_file = {
+            "configurable": {"thread_id": graph_thread_id},
+            "recursion_limit": 50,  # Default, make configurable if needed
+        }
+
+        try:
+            graph_produced_sdif_path = await self._invoke_xlsx_graph(
+                temp_input_file_for_graph, graph_config_for_file
+            )
+
+            target_intermediate_sdif_path = (
+                intermediate_sdif_dir
+                / f"intermediate_{input_xlsx_file.stem}_{graph_thread_id}.sdif"
+            )
+            shutil.move(
+                str(graph_produced_sdif_path),
+                str(target_intermediate_sdif_path),
+            )
+            logger.info(
+                f"Moved graph output for {input_xlsx_file.name} to {target_intermediate_sdif_path}"
+            )
+            return target_intermediate_sdif_path
+        except Exception as e:
+            error_msg = f"Failed to process file {input_xlsx_file.name} (using copy {temp_input_file_for_graph.name}) with xlsx-to-sdif graph: {e}"
+            logger.error(error_msg, exc_info=True)
+            # Re-raise to be caught by the main standardize method's loop or error handling
+            raise RuntimeError(
+                f"Error processing {input_xlsx_file.name}. Halting batch."
+            ) from e
+
+    async def _consolidate_intermediate_sdifs(
+        self, intermediate_sdif_paths: List[Path], final_output_path: Path
+    ) -> None:
+        """Consolidates intermediate SDIF files into the final output path."""
+        if not intermediate_sdif_paths:
+            # This case should ideally be handled before calling, but as a safeguard:
+            raise RuntimeError(
+                "No intermediate SDIF files were provided for consolidation."
+            )
+
+        if len(intermediate_sdif_paths) == 1:
+            logger.info(
+                f"Only one intermediate SDIF generated. Moving {intermediate_sdif_paths[0]} to {final_output_path}"
+            )
+            shutil.move(str(intermediate_sdif_paths[0]), str(final_output_path))
+        else:
+            logger.info(
+                f"Merging {len(intermediate_sdif_paths)} intermediate SDIF files into {final_output_path}"
+            )
+            merge_sdif_files(
+                source_db_paths=intermediate_sdif_paths,
+                target_db_path=final_output_path,
+            )
+
+    async def standardize(
+        self,
+        datasource: Datasource,
+        output_path: SDIFPath,
+        *,
+        overwrite: bool = False,
+        config: Optional[Dict[str, Any]] = None,
+        **kwargs: Any,
+    ) -> StandardizationResult:
+        """
+        Standardizes one or more XLSX files into a single SDIF file.
+
+        Args:
+            datasource: A single file path (str or Path) or a list of file paths
+                        to XLSX files.
+            output_path: The path where the final consolidated SDIF file will be saved.
+            overwrite: If True, overwrite the output_path if it already exists.
+                       Defaults to False.
+            config: General configuration options (currently not used by this standardizer
+                    for graph interaction but preserved for API consistency).
+            **kwargs: Additional keyword arguments (currently ignored).
+
+        Returns:
+            A StandardizationResult object containing the path to the final SDIF file.
+
+        Raises:
+            ValueError: If the datasource is invalid or no XLSX files are found.
+            RuntimeError: If critical errors occur during processing, such as the
+                          `xlsx-to-sdif` graph not being available or failing.
+            FileNotFoundError: If input files are not found or graph outputs are invalid.
+            FileExistsError: If output_path exists and overwrite is False.
+        """
+        if not xlsx_graph or not XLSXState:
+            raise RuntimeError(
+                "AIXLSXStandardizer cannot operate because `xlsx_to_sdif.graph` or `xlsx_to_sdif.state` is not available. "
+                "Please ensure the 'xlsx-to-sdif' library is installed and accessible."
+            )
+
+        xlsx_input_files = await self._resolve_and_filter_input_files(datasource)
+        final_output_path = self._prepare_final_output_path(output_path, overwrite)
+        run_temp_dir, intermediate_sdif_dir, temp_input_copies_dir = (
+            self._setup_temp_directories()
+        )
+
+        intermediate_sdif_paths: List[Path] = []
+        processing_errors: List[str] = []
+
+        try:
+            # Process each file sequentially. Consider asyncio.gather for parallel if graph supports it well for many files.
+            for i, input_xlsx_file in enumerate(xlsx_input_files):
+                try:
+                    logger.info(
+                        f"Starting processing for file {i + 1}/{len(xlsx_input_files)}: {input_xlsx_file.name}"
+                    )
+                    intermediate_sdif_path = (
+                        await self._process_single_file_to_intermediate_sdif(
+                            input_xlsx_file,
+                            final_output_path.stem,  # Pass stem for unique naming
+                            temp_input_copies_dir,
+                            intermediate_sdif_dir,
+                        )
+                    )
+                    intermediate_sdif_paths.append(intermediate_sdif_path)
+                except Exception:
+                    logger.error(
+                        f"Halting standardization due to error processing {input_xlsx_file.name}."
+                    )
+                    raise  # Re-raise the exception to be caught by the outer try/finally
+
+            if not intermediate_sdif_paths:
+                # This condition might be redundant if _process_single_file_to_intermediate_sdif always raises on failure
+                # and we re-raise immediately.
+                if processing_errors:  # This list would be empty if we fail fast
+                    raise RuntimeError(
+                        f"No XLSX files were successfully processed. Errors: {'; '.join(processing_errors)}"
+                    )
+                else:
+                    raise RuntimeError(
+                        "No intermediate SDIF files were generated, though no specific errors were caught."
+                    )
+
+            await self._consolidate_intermediate_sdifs(
+                intermediate_sdif_paths, final_output_path
+            )
+
+            logger.info(f"Successfully created final SDIF: {final_output_path}")
+            return StandardizationResult(
+                output_path=final_output_path, file_configs=None
+            )  # file_configs not available from this process
+
+        finally:
+            if run_temp_dir.exists():
+                try:
+                    shutil.rmtree(run_temp_dir)
+                    logger.debug(f"Cleaned up temporary directory: {run_temp_dir}")
+                except Exception as e_clean:
+                    logger.error(
+                        f"Error cleaning up temporary directory {run_temp_dir}: {e_clean}",
+                        exc_info=True,
+                    )

satif_ai/transform.py

@@ -1,7 +1,7 @@
 from pathlib import Path
 from typing import Any, Dict, List, Optional
 
-from fastmcp import FastMCP
+from fastmcp import Client, FastMCP
 from fastmcp.client.transports import FastMCPTransport
 from satif_core.code_executors.base import CodeExecutor
 from satif_core.transformation_builders.base import AsyncTransformationBuilder
@@ -28,7 +28,7 @@ async def atransform(
     transformation_builder: Optional[AsyncTransformationBuilder] = None,
     code_executor: Optional[CodeExecutor] = None,
     mcp_server: Optional[FastMCP] = None,
-    mcp_transport: Optional[FastMCPTransport] = None,
+    mcp_client: Optional[Client] = None,
     llm_model: str = "o4-mini",
     schema_only: bool = False,
     representer_kwargs: Optional[Dict[str, Any]] = None,
@@ -38,7 +38,7 @@ async def atransform(
     an AI-generated or provided transformation code.
 
     This function orchestrates the process of:
-    1. Optionally generating transformation code using an AI model via a `CodeBuilder`
+    1. Optionally generating transformation code using an AI model via a `TransformationBuilder`
        if `transformation_code` is not provided.
        explicitly passed.
     2. Executing the transformation code using a `CodeTransformer` and a `CodeExecutor`.
@@ -53,18 +53,21 @@ async def atransform(
                              output filenames if the transformation result keys match.
         instructions: Optional. Natural language instructions for the AI to generate
                       the transformation code. Used if `transformation_code` is None.
+        output_path: Path to the directory where transformation outputs will be saved.
         transformation_code: Optional. Pre-existing Python code for the transformation.
                              If None, code will be generated by the `transformation_builder`.
         transformation_builder: Optional. An `AsyncTransformationBuilder` instance responsible for generating
                       the transformation code if `transformation_code` is not provided.
-                      If None, a `TransformationAsyncCodeBuilder` is instantiated.
+                      If None, a `SyncpulseTransformationBuilder` is instantiated.
         code_executor: Optional. A `CodeExecutor` instance for running the transformation
                        code. If None, a `LocalCodeExecutor` is used.
         mcp_server: Optional. A `FastMCP` server instance for the AI code builder.
-                    Defaults to the global `mcp` instance if `transformation_builder` is None.
-        mcp_transport: Optional. A `FastMCPTransport` instance for communication with
-                       the `mcp_server`. Defaults to a new transport using `mcp_server`
-                       if `transformation_builder` is None.
+                    Defaults to the global `mcp` instance if `transformation_builder` is None and
+                    a new `SyncpulseTransformationBuilder` is being created.
+        mcp_client: Optional. A user-provided `Client` instance. If provided when
+                    `transformation_builder` is None, it will be used by the internally
+                    created `SyncpulseTransformationBuilder`. The caller is responsible for
+                    managing the lifecycle of a provided client.
         llm_model: The language model to use for code generation (e.g., "o4-mini").
                    Used if `transformation_builder` is None.
         schema_only: If True, the transformation aims to match only the schema (headers)
@@ -78,44 +81,75 @@ async def atransform(
         A `TransformationResult` object containing the path to the output
         and the transformation code used.
     """
-    if transformation_builder is None:
-        if mcp_server is None:
-            mcp_server = mcp
-
-        if mcp_transport is None:
-            mcp_transport = FastMCPTransport(mcp=mcp_server)
-
-        openai_compatible_mcp = OpenAICompatibleMCP(mcp=mcp_server)
-        await openai_compatible_mcp.connect()
-
-        transformation_builder = SyncpulseTransformationBuilder(
-            mcp_server=openai_compatible_mcp,
-            mcp_session=mcp_transport,
-            llm_model=llm_model,
-        )
-
-    if transformation_code is None:
-        function_code = await transformation_builder.build(
-            sdif=sdif,
-            output_target_files=output_target_files,
-            instructions=instructions,
-            schema_only=schema_only,
-            representer_kwargs=representer_kwargs,
-        )
-    else:
-        function_code = transformation_code
-
-    if code_executor is None:
-        code_executor = LocalCodeExecutor()
+    current_transformation_code: Optional[str] = transformation_code
+    active_builder: Optional[AsyncTransformationBuilder] = transformation_builder
+
+    _openai_mcp_instance: Optional[OpenAICompatibleMCP] = None
+    openai_mcp_managed_locally = False
+
+    # If code isn't provided, we need a builder. If a builder isn't provided, we create one.
+    if current_transformation_code is None:
+        if active_builder is None:
+            # Create SyncpulseTransformationBuilder
+            _effective_mcp_server = mcp_server if mcp_server is not None else mcp
+
+            _openai_mcp_instance = OpenAICompatibleMCP(mcp=_effective_mcp_server)
+            await _openai_mcp_instance.connect()
+            openai_mcp_managed_locally = True
+
+            if mcp_client is None:  # No user-provided client, create and manage one
+                mcp_transport = FastMCPTransport(mcp=_effective_mcp_server)
+                async with Client(mcp_transport) as new_client:
+                    active_builder = SyncpulseTransformationBuilder(
+                        mcp_server=_openai_mcp_instance,
+                        mcp_session=new_client.session,
+                        llm_model=llm_model,
+                    )
+                    current_transformation_code = await active_builder.build(
+                        sdif=sdif,
+                        output_target_files=output_target_files,
+                        instructions=instructions,
+                        schema_only=schema_only,
+                        representer_kwargs=representer_kwargs,
+                    )
+            else:
+                active_builder = SyncpulseTransformationBuilder(
+                    mcp_server=_openai_mcp_instance,
+                    mcp_session=mcp_client,  # Use the provided client
+                    llm_model=llm_model,
+                )
+                current_transformation_code = await active_builder.build(
+                    sdif=sdif,
+                    output_target_files=output_target_files,
+                    instructions=instructions,
+                    schema_only=schema_only,
+                    representer_kwargs=representer_kwargs,
+                )
+
+    # Disconnect OpenAICompatibleMCP if it was created and connected locally
+    if (
+        openai_mcp_managed_locally
+        and _openai_mcp_instance
+        and _openai_mcp_instance._is_connected
+    ):
+        await _openai_mcp_instance.cleanup()
+
+    if current_transformation_code is None:
+        raise ValueError("Transformation code could not be obtained or generated.")
+
+    # Code Executor and Transformation
+    _code_executor = code_executor if code_executor is not None else LocalCodeExecutor()
 
     transformer = CodeTransformer(
-        function=function_code,
-        code_executor=code_executor,
+        function=current_transformation_code,
+        code_executor=_code_executor,
     )
 
-    output_path = transformer.export(
+    exported_artifact_path = transformer.export(
         sdif=sdif,
         output_path=output_path,
     )
 
-    return TransformationResult(output_path=output_path, function_code=function_code)
+    return TransformationResult(
+        output_path=exported_artifact_path, function_code=current_transformation_code
+    )