fast-agent-mcp 0.2.48__tar.gz → 0.2.50__tar.gz

{fast_agent_mcp-0.2.48 → fast_agent_mcp-0.2.50}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fast-agent-mcp
-Version: 0.2.48
+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-0.2.48 → fast_agent_mcp-0.2.50}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "fast-agent-mcp"
-version = "0.2.48"
+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-0.2.48 → fast_agent_mcp-0.2.50}/src/mcp_agent/config.py

@@ -8,6 +8,7 @@ import re
 from pathlib import Path
 from typing import Any, Dict, List, Literal, Optional, Tuple
 
+from mcp import Implementation
 from pydantic import BaseModel, ConfigDict, field_validator
 from pydantic_settings import BaseSettings, SettingsConfigDict
 
@@ -61,11 +62,9 @@ class MCPServerSettings(BaseModel):
     Represents the configuration for an individual server.
     """
 
-    # TODO: saqadri - server name should be something a server can provide itself during initialization
     name: str | None = None
     """The name of the server."""
 
-    # TODO: saqadri - server description should be something a server can provide itself during initialization
     description: str | None = None
     """The description of the server."""
 
@@ -108,6 +107,8 @@ class MCPServerSettings(BaseModel):
     cwd: str | None = None
     """Working directory for the executed server command."""
 
+    implementation: Implementation | None = None
+
 
 class MCPSettings(BaseModel):
     """Configuration for all MCP servers."""

{fast_agent_mcp-0.2.48 → 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-0.2.48 → 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-0.2.48 → fast_agent_mcp-0.2.50}/src/mcp_agent/llm/model_factory.py

@@ -143,6 +143,7 @@ class ModelFactory:
         "gemini2": "gemini-2.0-flash",
         "gemini25": "gemini-2.5-flash-preview-05-20",
         "gemini25pro": "gemini-2.5-pro-preview-05-06",
+        "kimi": "groq.moonshotai/kimi-k2-instruct",
     }
 
     # Mapping of providers to their LLM classes

{fast_agent_mcp-0.2.48 → 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-0.2.48 → fast_agent_mcp-0.2.50}/src/mcp_agent/llm/providers/augmented_llm_azure.py

@@ -18,7 +18,7 @@ def _extract_resource_name(url: str) -> str | None:
     return host.replace(suffix, "") if host.endswith(suffix) else None
 
 
-DEFAULT_AZURE_API_VERSION = "2023-05-15"
+DEFAULT_AZURE_API_VERSION = "2024-10-21"
 
 
 class AzureOpenAIAugmentedLLM(OpenAIAugmentedLLM):

{fast_agent_mcp-0.2.48 → 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-0.2.48 → 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
@@ -412,6 +415,7 @@ class OpenAIAugmentedLLM(AugmentedLLM[ChatCompletionMessageParam, ChatCompletion
                     )
 
                 tool_results = []
+
                 for tool_call in message.tool_calls:
                     self.show_tool_call(
                         available_tools,
@@ -428,12 +432,24 @@ class OpenAIAugmentedLLM(AugmentedLLM[ChatCompletionMessageParam, ChatCompletion
                             else from_json(tool_call.function.arguments, allow_partial=True),
                         ),
                     )
-                    result = await self.call_tool(tool_call_request, tool_call.id)
-                    self.show_tool_result(result)
 
-                    tool_results.append((tool_call.id, result))
-                    responses.extend(result.content)
-                messages.extend(OpenAIConverter.convert_function_results_to_openai(tool_results))
+                    try:
+                        result = await self.call_tool(tool_call_request, tool_call.id)
+                        self.show_tool_result(result)
+                        tool_results.append((tool_call.id, result))
+                        responses.extend(result.content)
+                    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)}")]
+                        )
+                        tool_results.append((tool_call.id, error_result))
+
+                converted_messages = OpenAIConverter.convert_function_results_to_openai(
+                    tool_results
+                )
+                messages.extend(converted_messages)
 
                 self.logger.debug(
                     f"Iteration {i}: Tool call results: {str(tool_results) if tool_results else 'None'}"

{fast_agent_mcp-0.2.48 → 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-0.2.48 → 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-0.2.48 → fast_agent_mcp-0.2.50}/src/mcp_agent/llm/providers/multipart_converter_openai.py

@@ -441,9 +441,6 @@ class OpenAIConverter:
         # Convert to OpenAI format
         user_message = OpenAIConverter.convert_to_openai(non_text_multipart)
 
-        # We need to add tool_call_id manually
-        user_message["tool_call_id"] = tool_call_id
-
         return (tool_message, [user_message])
 
     @staticmethod
@@ -461,22 +458,42 @@ class OpenAIConverter:
         Returns:
             List of OpenAI API messages for tool responses
         """
-        messages = []
+        tool_messages = []
+        user_messages = []
+        has_mixed_content = False
 
         for tool_call_id, result in results:
-            converted = OpenAIConverter.convert_tool_result_to_openai(
-                tool_result=result,
-                tool_call_id=tool_call_id,
-                concatenate_text_blocks=concatenate_text_blocks,
-            )
-
-            # Handle the case where we have mixed content and get back a tuple
-            if isinstance(converted, tuple):
-                tool_message, additional_messages = converted
-                messages.append(tool_message)
-                messages.extend(additional_messages)
-            else:
-                # Single message case (text-only)
-                messages.append(converted)
+            try:
+                converted = OpenAIConverter.convert_tool_result_to_openai(
+                    tool_result=result,
+                    tool_call_id=tool_call_id,
+                    concatenate_text_blocks=concatenate_text_blocks,
+                )
 
+                # Handle the case where we have mixed content and get back a tuple
+                if isinstance(converted, tuple):
+                    tool_message, additional_messages = converted
+                    tool_messages.append(tool_message)
+                    user_messages.extend(additional_messages)
+                    has_mixed_content = True
+                else:
+                    # Single message case (text-only)
+                    tool_messages.append(converted)
+            except Exception as e:
+                _logger.error(f"Failed to convert tool_call_id={tool_call_id}: {e}")
+                # Create a basic tool response to prevent missing tool_call_id error
+                fallback_message = {
+                    "role": "tool",
+                    "tool_call_id": tool_call_id,
+                    "content": f"[Conversion error: {str(e)}]",
+                }
+                tool_messages.append(fallback_message)
+
+        # CONDITIONAL REORDERING: Only reorder if there are user messages (mixed content)
+        if has_mixed_content and user_messages:
+            # Reorder: All tool messages first (OpenAI sequence), then user messages (vision context)
+            messages = tool_messages + user_messages
+        else:
+            # Pure tool responses - keep original order to preserve context (snapshots, etc.)
+            messages = tool_messages
         return messages

{fast_agent_mcp-0.2.48 → fast_agent_mcp-0.2.50}/src/mcp_agent/logging/rich_progress.py

@@ -21,7 +21,7 @@ class RichProgressDisplay:
         self._progress = Progress(
             SpinnerColumn(spinner_name="simpleDotsScrolling"),
             TextColumn(
-                "[progress.description]{task.description}|",
+                "[progress.description]{task.description}▎",
                 #                table_column=Column(max_width=16),
             ),
             TextColumn(text_format="{task.fields[target]:<16}", style="Bold Blue"),
@@ -77,11 +77,12 @@ class RichProgressDisplay:
             ProgressAction.LOADED: "dim green",
             ProgressAction.INITIALIZED: "dim green",
             ProgressAction.CHATTING: "bold blue",
-            ProgressAction.STREAMING: "bold blue",  # Same color as chatting
+            ProgressAction.STREAMING: "bold green",  # Assistant Colour
             ProgressAction.ROUTING: "bold blue",
             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",
@@ -107,18 +108,49 @@ class RichProgressDisplay:
         # Ensure no None values in the update
         # For streaming, use custom description immediately to avoid flashing
         if event.action == ProgressAction.STREAMING and event.streaming_tokens:
-            formatted_tokens = f"↓ {event.streaming_tokens.strip()}".ljust(15)
+            # Account for [dim][/dim] tags (11 characters) in padding calculation
+            formatted_tokens = f"▎[dim]◀[/dim] {event.streaming_tokens.strip()}".ljust(17 + 11)
             description = f"[{self._get_action_style(event.action)}]{formatted_tokens}"
+        elif event.action == ProgressAction.CHATTING:
+            # Add special formatting for chatting with dimmed arrow
+            formatted_text = f"▎[dim]▶[/dim] {event.action.value.strip()}".ljust(17 + 11)
+            description = f"[{self._get_action_style(event.action)}]{formatted_text}"
+        elif event.action == ProgressAction.CALLING_TOOL:
+            # 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,
-        )
+            description = f"[{self._get_action_style(event.action)}]▎ {event.action.value:<15}"
+
+        # 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