fast-agent-mcp 0.2.49__tar.gz → 0.2.50__tar.gz

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fast-agent-mcp
-Version: 0.2.49
+Version: 0.2.50
 Summary: Define, Prompt and Test MCP enabled Agents and Workflows
 Author-email: Shaun Smith <fastagent@llmindset.co.uk>
 License:                                  Apache License
@@ -208,8 +208,8 @@ License-File: LICENSE
 Classifier: License :: OSI Approved :: Apache Software License
 Classifier: Operating System :: OS Independent
 Classifier: Programming Language :: Python :: 3
-Requires-Python: >=3.12
-Requires-Dist: a2a-sdk>=0.2.9
+Requires-Python: >=3.13
+Requires-Dist: a2a-sdk>=0.3.0
 Requires-Dist: aiohttp>=3.11.13
 Requires-Dist: anthropic>=0.59.0
 Requires-Dist: azure-identity>=1.14.0
@@ -232,7 +232,7 @@ Requires-Dist: pydantic>=2.10.4
 Requires-Dist: pyperclip>=1.9.0
 Requires-Dist: pyyaml>=6.0.2
 Requires-Dist: rich>=14.1.0
-Requires-Dist: tensorzero>=2025.6.3
+Requires-Dist: tensorzero>=2025.7.5
 Requires-Dist: typer>=0.15.1
 Provides-Extra: azure
 Requires-Dist: azure-identity>=1.14.0; extra == 'azure'

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "fast-agent-mcp"
-version = "0.2.49"
+version = "0.2.50"
 description = "Define, Prompt and Test MCP enabled Agents and Workflows"
 readme = "README.md"
 license = { file = "LICENSE" }
@@ -12,7 +12,7 @@ classifiers = [
     "License :: OSI Approved :: Apache Software License",
     "Operating System :: OS Independent"
 ]
-requires-python = ">=3.12"
+requires-python = ">=3.13"
 dependencies = [
     "fastapi>=0.115.6",
     "mcp==1.12.1",
@@ -34,9 +34,9 @@ dependencies = [
     "opentelemetry-instrumentation-mcp>=0.43.1; python_version >= '3.10' and python_version < '4.0'",
     "google-genai>=1.27.0",
     "opentelemetry-instrumentation-google-genai>=0.3b0",
-    "tensorzero>=2025.6.3",
+    "tensorzero>=2025.7.5",
     "deprecated>=1.2.18",
-    "a2a-sdk>=0.2.9",
+    "a2a-sdk>=0.3.0",
     "email-validator>=2.2.0",
     "pyperclip>=1.9.0",
 ]

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/src/mcp_agent/event_progress.py

@@ -20,6 +20,7 @@ class ProgressAction(str, Enum):
     PLANNING = "Planning"
     READY = "Ready"
     CALLING_TOOL = "Calling Tool"
+    TOOL_PROGRESS = "Tool Progress"
     UPDATED = "Updated"
     FINISHED = "Finished"
     SHUTDOWN = "Shutdown"
@@ -35,6 +36,8 @@ class ProgressEvent(BaseModel):
     details: Optional[str] = None
     agent_name: Optional[str] = None
     streaming_tokens: Optional[str] = None  # Special field for streaming token count
+    progress: Optional[float] = None  # Current progress value
+    total: Optional[float] = None  # Total value for progress calculation
 
     def __str__(self) -> str:
         """Format the progress event for display."""
@@ -86,6 +89,12 @@ def convert_log_event(event: Event) -> Optional[ProgressEvent]:
             details = f"{server_name} ({tool_name})"
         else:
             details = f"{server_name}"
+        
+        # For TOOL_PROGRESS, use progress message if available, otherwise keep default
+        if progress_action == ProgressAction.TOOL_PROGRESS:
+            progress_message = event_data.get("details", "")
+            if progress_message:  # Only override if message is non-empty
+                details = progress_message
 
     elif "augmented_llm" in namespace:
         model = event_data.get("model", "")
@@ -104,10 +113,19 @@ def convert_log_event(event: Event) -> Optional[ProgressEvent]:
     if progress_action == ProgressAction.STREAMING:
         streaming_tokens = event_data.get("details", "")
     
+    # Extract progress data for TOOL_PROGRESS actions
+    progress = None
+    total = None
+    if progress_action == ProgressAction.TOOL_PROGRESS:
+        progress = event_data.get("progress")
+        total = event_data.get("total")
+    
     return ProgressEvent(
         action=ProgressAction(progress_action),
         target=target or "unknown",
         details=details,
         agent_name=event_data.get("agent_name"),
         streaming_tokens=streaming_tokens,
+        progress=progress,
+        total=total,
     )

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/src/mcp_agent/llm/model_database.py

@@ -22,6 +22,12 @@ class ModelParameters(BaseModel):
     tokenizes: List[str]
     """List of supported content types for tokenization"""
 
+    json_mode: None | str = "schema"
+    """Structured output style. 'schema', 'object' or None for unsupported """
+
+    reasoning: None | str = None
+    """Reasoning output style. 'tags' if enclosed in <thinking> tags, 'none' if not used"""
+
 
 class ModelDatabase:
     """Centralized model configuration database"""
@@ -87,6 +93,13 @@ class ModelDatabase:
     QWEN_STANDARD = ModelParameters(
         context_window=32000, max_output_tokens=8192, tokenizes=QWEN_MULTIMODAL
     )
+    QWEN3_REASONER = ModelParameters(
+        context_window=131072,
+        max_output_tokens=16384,
+        tokenizes=TEXT_ONLY,
+        json_mode="object",
+        reasoning="tags",
+    )
 
     FAST_AGENT_STANDARD = ModelParameters(
         context_window=1000000, max_output_tokens=100000, tokenizes=TEXT_ONLY
@@ -125,6 +138,13 @@ class ModelDatabase:
         context_window=65536, max_output_tokens=32768, tokenizes=TEXT_ONLY
     )
 
+    DEEPSEEK_DISTILL = ModelParameters(
+        context_window=131072,
+        max_output_tokens=131072,
+        tokenizes=TEXT_ONLY,
+        json_mode="object",
+        reasoning="tags",
+    )
     GEMINI_2_5_PRO = ModelParameters(
         context_window=2097152, max_output_tokens=8192, tokenizes=GOOGLE_MULTIMODAL
     )
@@ -214,6 +234,8 @@ class ModelDatabase:
         "grok-3-fast": GROK_3,
         "grok-3-mini-fast": GROK_3,
         "moonshotai/kimi-k2-instruct": KIMI_MOONSHOT,
+        "qwen/qwen3-32b": QWEN3_REASONER,
+        "deepseek-r1-distill-llama-70b": DEEPSEEK_DISTILL,
     }
 
     @classmethod
@@ -239,6 +261,18 @@ class ModelDatabase:
         params = cls.get_model_params(model)
         return params.tokenizes if params else None
 
+    @classmethod
+    def get_json_mode(cls, model: str) -> str | None:
+        """Get supported json mode (structured output) for a model"""
+        params = cls.get_model_params(model)
+        return params.json_mode if params else None
+
+    @classmethod
+    def get_reasoning(cls, model: str) -> str | None:
+        """Get supported reasoning output style for a model"""
+        params = cls.get_model_params(model)
+        return params.reasoning if params else None
+
     @classmethod
     def get_default_max_tokens(cls, model: str) -> int:
         """Get default max_tokens for RequestParams based on model"""

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/src/mcp_agent/llm/providers/augmented_llm_aliyun.py

@@ -12,15 +12,14 @@ class AliyunAugmentedLLM(OpenAIAugmentedLLM):
 
     def _initialize_default_params(self, kwargs: dict) -> RequestParams:
         """Initialize Aliyun-specific default parameters"""
+        # Get base defaults from parent (includes ModelDatabase lookup)
+        base_params = super()._initialize_default_params(kwargs)
+        
+        # Override with Aliyun-specific settings
         chosen_model = kwargs.get("model", DEFAULT_QWEN_MODEL)
-
-        return RequestParams(
-            model=chosen_model,
-            systemPrompt=self.instruction,
-            parallel_tool_calls=True,
-            max_iterations=20,
-            use_history=True,
-        )
+        base_params.model = chosen_model
+        
+        return base_params
 
     def _base_url(self) -> str:
         base_url = None

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/src/mcp_agent/llm/providers/augmented_llm_deepseek.py

@@ -22,15 +22,14 @@ class DeepSeekAugmentedLLM(OpenAIAugmentedLLM):
 
     def _initialize_default_params(self, kwargs: dict) -> RequestParams:
         """Initialize Deepseek-specific default parameters"""
+        # Get base defaults from parent (includes ModelDatabase lookup)
+        base_params = super()._initialize_default_params(kwargs)
+        
+        # Override with Deepseek-specific settings
         chosen_model = kwargs.get("model", DEFAULT_DEEPSEEK_MODEL)
-
-        return RequestParams(
-            model=chosen_model,
-            systemPrompt=self.instruction,
-            parallel_tool_calls=True,
-            max_iterations=20,
-            use_history=True,
-        )
+        base_params.model = chosen_model
+        
+        return base_params
 
     def _base_url(self) -> str:
         base_url = None

fast_agent_mcp-0.2.50/src/mcp_agent/llm/providers/augmented_llm_groq.py

@@ -0,0 +1,103 @@
+from typing import List, Tuple, Type, cast
+
+from pydantic_core import from_json
+
+from mcp_agent.core.request_params import RequestParams
+from mcp_agent.llm.model_database import ModelDatabase
+from mcp_agent.llm.provider_types import Provider
+from mcp_agent.llm.providers.augmented_llm_openai import OpenAIAugmentedLLM
+from mcp_agent.logging.logger import get_logger
+from mcp_agent.mcp.helpers.content_helpers import get_text, split_thinking_content
+from mcp_agent.mcp.interfaces import ModelT
+from mcp_agent.mcp.prompt_message_multipart import PromptMessageMultipart
+
+GROQ_BASE_URL = "https://api.groq.com/openai/v1"
+DEFAULT_GROQ_MODEL = "moonshotai/kimi-k2-instruct"
+
+### There is some big refactorings to be had quite easily here now:
+### - combining the structured output type handling
+### - deduplicating between this and the deepseek llm
+
+
+class GroqAugmentedLLM(OpenAIAugmentedLLM):
+    def __init__(self, *args, **kwargs) -> None:
+        super().__init__(*args, provider=Provider.GROQ, **kwargs)
+
+    def _initialize_default_params(self, kwargs: dict) -> RequestParams:
+        """Initialize Groq default parameters"""
+        # Get base defaults from parent (includes ModelDatabase lookup)
+        base_params = super()._initialize_default_params(kwargs)
+
+        # Override with Groq-specific settings
+        chosen_model = kwargs.get("model", DEFAULT_GROQ_MODEL)
+        base_params.model = chosen_model
+        base_params.parallel_tool_calls = False
+
+        return base_params
+
+    async def _apply_prompt_provider_specific_structured(
+        self,
+        multipart_messages: List[PromptMessageMultipart],
+        model: Type[ModelT],
+        request_params: RequestParams | None = None,
+    ) -> Tuple[ModelT | None, PromptMessageMultipart]:  # noqa: F821
+        request_params = self.get_request_params(request_params)
+
+        assert self.default_request_params
+        llm_model = self.default_request_params.model or DEFAULT_GROQ_MODEL
+        json_mode: str | None = ModelDatabase.get_json_mode(llm_model)
+        if "json_object" == json_mode:
+            request_params.response_format = {"type": "json_object"}
+
+        # Get the full schema and extract just the properties
+        full_schema = model.model_json_schema()
+        properties = full_schema.get("properties", {})
+        required_fields = full_schema.get("required", [])
+
+        # Create a cleaner format description
+        format_description = "{\n"
+        for field_name, field_info in properties.items():
+            field_type = field_info.get("type", "string")
+            description = field_info.get("description", "")
+            format_description += f'  "{field_name}": "{field_type}"'
+            if description:
+                format_description += f"  // {description}"
+            if field_name in required_fields:
+                format_description += "  // REQUIRED"
+            format_description += "\n"
+        format_description += "}"
+
+        multipart_messages[-1].add_text(
+            f"""YOU MUST RESPOND WITH A JSON OBJECT IN EXACTLY THIS FORMAT:
+            {format_description}
+
+            IMPORTANT RULES:
+            - Respond ONLY with the JSON object, no other text
+            - Do NOT include "properties" or "schema" wrappers
+            - Do NOT use code fences or markdown
+            - The response must be valid JSON that matches the format above
+            - All required fields must be included"""
+        )
+
+        result: PromptMessageMultipart = await self._apply_prompt_provider_specific(
+            multipart_messages, request_params
+        )
+        reasoning_mode: str | None = ModelDatabase.get_reasoning(llm_model)
+        try:
+            text = get_text(result.content[-1]) or ""
+            if "tags" == reasoning_mode:
+                _, text = split_thinking_content(text)
+            json_data = from_json(text, allow_partial=True)
+            validated_model = model.model_validate(json_data)
+            return cast("ModelT", validated_model), result
+        except ValueError as e:
+            logger = get_logger(__name__)
+            logger.warning(f"Failed to parse structured response: {str(e)}")
+            return None, result
+
+    def _base_url(self) -> str:
+        base_url = None
+        if self.context.config and self.context.config.groq:
+            base_url = self.context.config.groq.base_url
+
+        return base_url if base_url else GROQ_BASE_URL

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/src/mcp_agent/llm/providers/augmented_llm_openai.py

@@ -307,6 +307,7 @@ class OpenAIAugmentedLLM(AugmentedLLM[ChatCompletionMessageParam, ChatCompletion
         request_params = self.get_request_params(request_params=request_params)
 
         responses: List[ContentBlock] = []
+        model_name = self.default_request_params.model or DEFAULT_OPENAI_MODEL
 
         # TODO -- move this in to agent context management / agent group handling
         messages: List[ChatCompletionMessageParam] = []
@@ -347,7 +348,6 @@ class OpenAIAugmentedLLM(AugmentedLLM[ChatCompletionMessageParam, ChatCompletion
             stream = await self._openai_client().chat.completions.create(**arguments)
             # Process the stream
             response = await self._process_stream(stream, self.default_request_params.model)
-
             # Track usage if response is valid and has usage data
             if (
                 hasattr(response, "usage")
@@ -391,6 +391,9 @@ class OpenAIAugmentedLLM(AugmentedLLM[ChatCompletionMessageParam, ChatCompletion
             # Convert to dict and remove None values
             message_dict = message.model_dump()
             message_dict = {k: v for k, v in message_dict.items() if v is not None}
+            if model_name == "deepseek-r1-distill-llama-70b":
+                message_dict.pop("reasoning", None)
+
             messages.append(message_dict)
 
             message_text = message.content
@@ -412,9 +415,8 @@ class OpenAIAugmentedLLM(AugmentedLLM[ChatCompletionMessageParam, ChatCompletion
                     )
 
                 tool_results = []
-                
+
                 for tool_call in message.tool_calls:
-                    
                     self.show_tool_call(
                         available_tools,
                         tool_call.function.name,
@@ -430,7 +432,7 @@ class OpenAIAugmentedLLM(AugmentedLLM[ChatCompletionMessageParam, ChatCompletion
                             else from_json(tool_call.function.arguments, allow_partial=True),
                         ),
                     )
-                    
+
                     try:
                         result = await self.call_tool(tool_call_request, tool_call.id)
                         self.show_tool_result(result)
@@ -439,10 +441,14 @@ class OpenAIAugmentedLLM(AugmentedLLM[ChatCompletionMessageParam, ChatCompletion
                     except Exception as e:
                         self.logger.error(f"Tool call {tool_call.id} failed with error: {e}")
                         # Still add the tool_call_id with an error result to prevent missing responses
-                        error_result = CallToolResult(content=[TextContent(type="text", text=f"Tool call failed: {str(e)}")])
+                        error_result = CallToolResult(
+                            content=[TextContent(type="text", text=f"Tool call failed: {str(e)}")]
+                        )
                         tool_results.append((tool_call.id, error_result))
-                
-                converted_messages = OpenAIConverter.convert_function_results_to_openai(tool_results)
+
+                converted_messages = OpenAIConverter.convert_function_results_to_openai(
+                    tool_results
+                )
                 messages.extend(converted_messages)
 
                 self.logger.debug(

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/src/mcp_agent/llm/providers/augmented_llm_openrouter.py

@@ -17,24 +17,19 @@ class OpenRouterAugmentedLLM(OpenAIAugmentedLLM):
 
     def _initialize_default_params(self, kwargs: dict) -> RequestParams:
         """Initialize OpenRouter-specific default parameters."""
+        # Get base defaults from parent (includes ModelDatabase lookup)
+        base_params = super()._initialize_default_params(kwargs)
+        
+        # Override with OpenRouter-specific settings
         # OpenRouter model names include the provider, e.g., "google/gemini-flash-1.5"
         # The model should be passed in the 'model' kwarg during factory creation.
         chosen_model = kwargs.get("model", DEFAULT_OPENROUTER_MODEL)
-        if not chosen_model:
-            # Unlike Deepseek, OpenRouter *requires* a model path in the identifier.
-            # The factory should extract this before calling the constructor.
-            # We rely on the model being passed correctly via kwargs.
-            # If it's still None here, it indicates an issue upstream (factory or user input).
-            # However, the base class _get_model handles the error if model is None.
-            pass
-
-        return RequestParams(
-            model=chosen_model,  # Will be validated by base class
-            systemPrompt=self.instruction,
-            parallel_tool_calls=True,  # Default based on OpenAI provider
-            max_iterations=20,  # Default based on OpenAI provider
-            use_history=True,  # Default based on OpenAI provider
-        )
+        if chosen_model:
+            base_params.model = chosen_model
+        # If it's still None here, it indicates an issue upstream (factory or user input).
+        # However, the base class _get_model handles the error if model is None.
+        
+        return base_params
 
     def _base_url(self) -> str:
         """Retrieve the OpenRouter base URL from config or use the default."""

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/src/mcp_agent/llm/providers/augmented_llm_xai.py

@@ -16,15 +16,15 @@ class XAIAugmentedLLM(OpenAIAugmentedLLM):
 
     def _initialize_default_params(self, kwargs: dict) -> RequestParams:
         """Initialize xAI parameters"""
+        # Get base defaults from parent (includes ModelDatabase lookup)
+        base_params = super()._initialize_default_params(kwargs)
+        
+        # Override with xAI-specific settings
         chosen_model = kwargs.get("model", DEFAULT_XAI_MODEL)
-
-        return RequestParams(
-            model=chosen_model,
-            systemPrompt=self.instruction,
-            parallel_tool_calls=False,
-            max_iterations=20,
-            use_history=True,
-        )
+        base_params.model = chosen_model
+        base_params.parallel_tool_calls = False
+        
+        return base_params
 
     def _base_url(self) -> str:
         base_url = os.getenv("XAI_BASE_URL", XAI_BASE_URL)

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/src/mcp_agent/logging/rich_progress.py

@@ -82,6 +82,7 @@ class RichProgressDisplay:
             ProgressAction.PLANNING: "bold blue",
             ProgressAction.READY: "dim green",
             ProgressAction.CALLING_TOOL: "bold magenta",
+            ProgressAction.TOOL_PROGRESS: "bold magenta",
             ProgressAction.FINISHED: "black on green",
             ProgressAction.SHUTDOWN: "black on red",
             ProgressAction.AGGREGATOR_INITIALIZED: "bold green",
@@ -118,16 +119,38 @@ class RichProgressDisplay:
             # Add special formatting for calling tool with dimmed arrow
             formatted_text = f"▎[dim]◀[/dim] {event.action.value}".ljust(17 + 11)
             description = f"[{self._get_action_style(event.action)}]{formatted_text}"
+        elif event.action == ProgressAction.TOOL_PROGRESS:
+            # Format similar to streaming - show progress numbers
+            if event.progress is not None:
+                if event.total is not None:
+                    progress_display = f"{int(event.progress)}/{int(event.total)}"
+                else:
+                    progress_display = str(int(event.progress))
+            else:
+                progress_display = "Processing"
+            formatted_text = f"▎[dim]▶[/dim] {progress_display}".ljust(17 + 11)
+            description = f"[{self._get_action_style(event.action)}]{formatted_text}"
         else:
             description = f"[{self._get_action_style(event.action)}]▎ {event.action.value:<15}"
 
-        self._progress.update(
-            task_id,
-            description=description,
-            target=event.target or task_name,  # Use task_name as fallback for target
-            details=event.details or "",
-            task_name=task_name,
-        )
+        # Update basic task information
+        update_kwargs = {
+            "description": description,
+            "target": event.target or task_name,  # Use task_name as fallback for target
+            "details": event.details or "",
+            "task_name": task_name,
+        }
+        
+        # For TOOL_PROGRESS events, update progress if available
+        if event.action == ProgressAction.TOOL_PROGRESS and event.progress is not None:
+            if event.total is not None:
+                update_kwargs["completed"] = event.progress
+                update_kwargs["total"] = event.total
+            else:
+                # If no total, just show as indeterminate progress
+                self._progress.reset(task_id)
+        
+        self._progress.update(task_id, **update_kwargs)
 
         if (
             event.action == ProgressAction.INITIALIZED

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/src/mcp_agent/mcp/helpers/content_helpers.py

@@ -156,3 +156,32 @@ def get_resource_text(result: ReadResourceResult, index: int = 0) -> Optional[st
         return content.text
 
     return None
+
+
+def split_thinking_content(message: str) -> tuple[Optional[str], str]:
+    """
+    Split a message into thinking and content parts.
+
+    Extracts content between <thinking> tags and returns it along with the remaining content.
+
+    Args:
+        message: A string that may contain a <thinking>...</thinking> block followed by content
+
+    Returns:
+        A tuple of (thinking_content, main_content) where:
+        - thinking_content: The content inside <thinking> tags, or None if not found/parsing fails
+        - main_content: The content after the thinking block, or the entire message if no thinking block
+    """
+    import re
+
+    # Pattern to match <thinking>...</thinking> at the start of the message
+    pattern = r"^<think>(.*?)</think>\s*(.*)$"
+    match = re.match(pattern, message, re.DOTALL)
+
+    if match:
+        thinking_content = match.group(1).strip()
+        main_content = match.group(2).strip()
+        return (thinking_content, main_content)
+    else:
+        # No thinking block found or parsing failed
+        return (None, message)

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.50}/src/mcp_agent/mcp/mcp_aggregator.py

@@ -12,6 +12,7 @@ from typing import (
 
 from mcp import GetPromptResult, ReadResourceResult
 from mcp.client.session import ClientSession
+from mcp.shared.session import ProgressFnT
 from mcp.types import (
     CallToolResult,
     ListToolsResult,
@@ -136,6 +137,24 @@ class MCPAggregator(ContextDependent):
         # Lock for refreshing tools from a server
         self._refresh_lock = Lock()
 
+    def _create_progress_callback(self, server_name: str, tool_name: str) -> "ProgressFnT":
+        """Create a progress callback function for tool execution."""
+        async def progress_callback(progress: float, total: float | None, message: str | None) -> None:
+            """Handle progress notifications from MCP tool execution."""
+            logger.info(
+                "Tool progress update",
+                data={
+                    "progress_action": ProgressAction.TOOL_PROGRESS,
+                    "tool_name": tool_name,
+                    "server_name": server_name,
+                    "agent_name": self.agent_name,
+                    "progress": progress,
+                    "total": total,
+                    "details": message or "",  # Put the message in details column
+                },
+            )
+        return progress_callback
+
     async def close(self) -> None:
         """
         Close all persistent connections when the aggregator is deleted.
@@ -468,6 +487,7 @@ class MCPAggregator(ContextDependent):
         method_name: str,
         method_args: Dict[str, Any] = None,
         error_factory: Callable[[str], R] = None,
+        progress_callback: ProgressFnT | None = None,
     ) -> R:
         """
         Generic method to execute operations on a specific server.
@@ -479,6 +499,7 @@ class MCPAggregator(ContextDependent):
             method_name: Name of the method to call on the client session
             method_args: Arguments to pass to the method
             error_factory: Function to create an error return value if the operation fails
+            progress_callback: Optional progress callback for operations that support it
 
         Returns:
             Result from the operation or an error result
@@ -487,7 +508,12 @@ class MCPAggregator(ContextDependent):
         async def try_execute(client: ClientSession):
             try:
                 method = getattr(client, method_name)
-                return await method(**method_args)
+                # For call_tool method, check if we need to add progress_callback
+                if method_name == "call_tool" and progress_callback:
+                    # The call_tool method signature includes progress_callback parameter
+                    return await method(**method_args, progress_callback=progress_callback)
+                else:
+                    return await method(**method_args)
             except Exception as e:
                 error_msg = (
                     f"Failed to {method_name} '{operation_name}' on server '{server_name}': {e}"
@@ -597,6 +623,10 @@ class MCPAggregator(ContextDependent):
         with tracer.start_as_current_span(f"MCP Tool: {server_name}/{local_tool_name}"):
             trace.get_current_span().set_attribute("tool_name", local_tool_name)
             trace.get_current_span().set_attribute("server_name", server_name)
+            
+            # Create progress callback for this tool execution
+            progress_callback = self._create_progress_callback(server_name, local_tool_name)
+            
             return await self._execute_on_server(
                 server_name=server_name,
                 operation_type="tool",
@@ -609,6 +639,7 @@ class MCPAggregator(ContextDependent):
                 error_factory=lambda msg: CallToolResult(
                     isError=True, content=[TextContent(type="text", text=msg)]
                 ),
+                progress_callback=progress_callback,
             )
 
     async def get_prompt(

fast_agent_mcp-0.2.49/src/mcp_agent/llm/providers/augmented_llm_groq.py

@@ -1,30 +0,0 @@
-from mcp_agent.core.request_params import RequestParams
-from mcp_agent.llm.provider_types import Provider
-from mcp_agent.llm.providers.augmented_llm_openai import OpenAIAugmentedLLM
-
-GROQ_BASE_URL = "https://api.groq.com/openai/v1"
-DEFAULT_GROQ_MODEL = ""
-
-
-class GroqAugmentedLLM(OpenAIAugmentedLLM):
-    def __init__(self, *args, **kwargs) -> None:
-        super().__init__(*args, provider=Provider.GROQ, **kwargs)
-
-    def _initialize_default_params(self, kwargs: dict) -> RequestParams:
-        """Initialize Groq default parameters"""
-        chosen_model = kwargs.get("model", DEFAULT_GROQ_MODEL)
-
-        return RequestParams(
-            model=chosen_model,
-            systemPrompt=self.instruction,
-            parallel_tool_calls=False,
-            max_iterations=20,
-            use_history=True,
-        )
-
-    def _base_url(self) -> str:
-        base_url = None
-        if self.context.config and self.context.config.groq:
-            base_url = self.context.config.groq.base_url
-
-        return base_url if base_url else GROQ_BASE_URL