fast-agent-mcp 0.2.49__tar.gz → 0.2.51__tar.gz

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.51}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fast-agent-mcp
-Version: 0.2.49
+Version: 0.2.51
 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-0.2.51}/examples/tensorzero/docker-compose.yml

@@ -63,8 +63,9 @@ services:
 
   mcp-server:
     build:
-      context: ../..
-      dockerfile: examples/tensorzero/mcp_server/Dockerfile
+      context: ./mcp_server
+      dockerfile: Dockerfile
+    working_dir: /app
     volumes:
       - ./mcp_server:/app
     ports:

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.51}/examples/tensorzero/fastagent.config.yaml

@@ -2,6 +2,7 @@ default_model: haiku
 
 tensorzero:
   base_url: http://localhost:3000
+  api_key: verysecret
 
 logger:
   level: "info"

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.51}/examples/tensorzero/mcp_server/Dockerfile

@@ -1,5 +1,4 @@
 FROM python:3.12-slim
-
 WORKDIR /app
 
 RUN apt-get update && apt-get install -y curl wget && \
@@ -9,20 +8,18 @@ RUN apt-get update && apt-get install -y curl wget && \
 
 RUN pip install uv
 
-COPY pyproject.toml /app/
-COPY uv.lock /app/
-COPY LICENSE /app/
-COPY README.md /app/
+# Copy dependency files only
+COPY pyproject.toml /tmp/
+#COPY uv.lock /tmp/
 
+# Install dependencies
+WORKDIR /tmp
 RUN uv pip install --system .
 
-COPY examples/tensorzero/mcp_server/mcp_server.py /app/
-COPY examples/tensorzero/mcp_server/entrypoint.sh /app/entrypoint.sh
-
-RUN chmod +x /app/entrypoint.sh
+# Switch back to /app (this will be overridden by volume mount)
+WORKDIR /app
 
+# These files will come from the volume mount at runtime
 EXPOSE 8000
-
 ENTRYPOINT ["/app/entrypoint.sh"]
-
 CMD ["uvicorn", "mcp_server:app", "--host", "0.0.0.0", "--port", "8000"]

fast_agent_mcp-0.2.51/examples/tensorzero/mcp_server/pyproject.toml

@@ -0,0 +1,11 @@
+[project]
+name = "mcp-server"
+version = "0.1.0"
+description = "Add your description here"
+readme = "README.md"
+requires-python = ">=3.12"
+dependencies = [
+    "fastmcp>=2.10.6",
+    "starlette>=0.46.2",
+    "uvicorn>=0.34.3",
+]

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.51}/hatch_build.py

@@ -29,6 +29,7 @@ class CustomBuildHook(BuildHookInterface):
             # examples/mcp/state-transfer -> src/mcp_agent/resources/examples/mcp/state-transfer
             "examples/mcp/state-transfer": "src/mcp_agent/resources/examples/mcp/state-transfer",
             "examples/mcp/elicitations": "src/mcp_agent/resources/examples/mcp/elicitations",
+            "examples/tensorzero": "src/mcp_agent/resources/examples/tensorzero",
         }
 
         print("Fast-agent build: Copying examples to resources...")

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.51}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "fast-agent-mcp"
-version = "0.2.49"
+version = "0.2.51"
 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",
 ]
@@ -136,3 +136,6 @@ ignore=["E501"]
 # More lenient settings for existing code
 [tool.ruff.lint.per-file-ignores]
 "tests/*" = ["ANN"]  # Don't require type annotations in tests
+
+[tool.uv.workspace]
+members = ["examples/tensorzero/mcp_server"]

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.51}/src/mcp_agent/cli/commands/quickstart.py

@@ -76,6 +76,25 @@ EXAMPLE_TYPES = {
         ],
         "create_subdir": True,
     },
+    "tensorzero": {
+        "description": "A complete example showcasing the TensorZero integration.\n"
+        "Includes the T0 Gateway, an MCP server, an interactive agent, and \n"
+        "multi-modal functionality.",
+        "files": [
+            ".env.sample",
+            "Makefile",
+            "README.md",
+            "agent.py",
+            "docker-compose.yml",
+            "fastagent.config.yaml",
+            "image_demo.py",
+            "simple_agent.py",
+            "mcp_server/",
+            "demo_images/",
+            "tensorzero_config/"
+        ],
+        "create_subdir": True,
+    },
 }
 
 
@@ -225,6 +244,27 @@ def copy_example_files(example_type: str, target_dir: Path, force: bool = False)
     return created
 
 
+def copy_project_template(source_dir: Path, dest_dir: Path, console: Console, force: bool = False):
+    """
+    Recursively copies a project template directory.
+    This is a helper to handle project-based quickstarts like TensorZero.
+    """
+    if dest_dir.exists():
+        if force:
+            console.print(f"[yellow]--force specified. Removing existing directory: {dest_dir}[/yellow]")
+            shutil.rmtree(dest_dir)
+        else:
+            console.print(f"[bold yellow]Directory '{dest_dir.name}' already exists.[/bold yellow] Use --force to overwrite.")
+            return False
+
+    try:
+        shutil.copytree(source_dir, dest_dir)
+        return True
+    except Exception as e:
+        console.print(f"[red]Error copying project template: {e}[/red]")
+        return False
+
+
 def show_overview() -> None:
     """Display an overview of available examples in a nicely formatted table."""
     console.print("\n[bold cyan]fast-agent quickstarts[/bold cyan]")
@@ -397,6 +437,73 @@ def _show_completion_message(example_type: str, created: list[str]) -> None:
         console.print("\n[yellow]No files were created.[/yellow]")
 
 
+@app.command(name="tensorzero", help="Create the TensorZero integration example project.")
+def tensorzero(
+    directory: Path = typer.Argument(
+        Path("."),
+        help="Directory where the 'tensorzero' project folder will be created.",
+    ),
+    force: bool = typer.Option(False, "--force", "-f", help="Force overwrite if project directory exists"),
+):
+    """Create the TensorZero project example."""
+    console.print("[bold green]Setting up the TensorZero quickstart example...[/bold green]")
+
+    dest_project_dir = directory.resolve() / "tensorzero"
+
+    # --- Find Source Directory ---
+    from importlib.resources import files
+    try:
+        # This path MUST match the "to" path from hatch_build.py
+        source_dir = files("mcp_agent").joinpath("resources").joinpath("examples").joinpath("tensorzero")
+        if not source_dir.is_dir():
+            raise FileNotFoundError  # Fallback to dev mode if resource isn't a dir
+    except (ImportError, ModuleNotFoundError, FileNotFoundError):
+        console.print("[yellow]Package resources not found. Falling back to development mode.[/yellow]")
+        # This path is relative to the project root in a development environment
+        source_dir = Path(__file__).parent.parent.parent.parent / "examples" / "tensorzero"
+
+    if not source_dir.exists() or not source_dir.is_dir():
+        console.print(f"[red]Error: Source project directory not found at '{source_dir}'[/red]")
+        raise typer.Exit(1)
+
+    console.print(f"Source directory: [dim]{source_dir}[/dim]")
+    console.print(f"Destination: [dim]{dest_project_dir}[/dim]")
+
+    # --- Copy Project and Show Message ---
+    if copy_project_template(source_dir, dest_project_dir, console, force):
+        console.print(
+            f"\n[bold green]✅ Success![/bold green] Your TensorZero project has been created in: [cyan]{dest_project_dir}[/cyan]"
+        )
+        console.print("\n[bold yellow]Next Steps:[/bold yellow]")
+        console.print("\n1. [bold]Navigate to your new project directory:[/bold]")
+        console.print(f"   [cyan]cd {dest_project_dir.relative_to(Path.cwd())}[/cyan]")
+
+        console.print("\n2. [bold]Set up your API keys:[/bold]")
+        console.print("   [cyan]cp .env.sample .env[/cyan]")
+        console.print(
+            "   [dim]Then, open the new '.env' file and add your OpenAI or Anthropic API key.[/dim]"
+        )
+
+        console.print("\n3. [bold]Start the required services (TensorZero Gateway & MCP Server):[/bold]")
+        console.print("   [cyan]docker compose up --build -d[/cyan]")
+        console.print(
+            "   [dim](This builds and starts the necessary containers in the background)[/dim]"
+        )
+
+        console.print("\n4. [bold]Run the interactive agent:[/bold]")
+        console.print("   [cyan]make agent[/cyan]  (or `uv run agent.py`)")
+        console.print("\nEnjoy exploring the TensorZero integration with fast-agent! ✨")
+
+
+@app.command(name="t0", help="Alias for the TensorZero quickstart.", hidden=True)
+def t0_alias(
+    directory: Path = typer.Argument(Path("."), help="Directory for the 'tensorzero' project folder."),
+    force: bool = typer.Option(False, "--force", "-f", help="Force overwrite"),
+):
+    """Alias for the `tensorzero` command."""
+    tensorzero(directory, force)
+
+
 @app.callback(invoke_without_command=True)
 def main(ctx: typer.Context) -> None:
     """Quickstart applications for fast-agent."""

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.51}/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-0.2.51}/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
@@ -107,7 +120,9 @@ class ModelDatabase:
     OPENAI_O3_MINI_SERIES = ModelParameters(
         context_window=200000, max_output_tokens=100000, tokenizes=TEXT_ONLY
     )
-
+    OPENAI_GPT_OSS_SERIES = ModelParameters(
+        context_window=131072, max_output_tokens=32766, tokenizes=TEXT_ONLY, json_mode="object"
+    )
     # TODO update to 32000
     ANTHROPIC_OPUS_4_VERSIONED = ModelParameters(
         context_window=200000, max_output_tokens=32000, tokenizes=ANTHROPIC_MULTIMODAL
@@ -125,6 +140,13 @@ 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 +236,10 @@ 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,
+        "openai/gpt-oss-120b": OPENAI_GPT_OSS_SERIES,
+        "openai/gpt-oss-20b": OPENAI_GPT_OSS_SERIES,
     }
 
     @classmethod
@@ -239,6 +265,18 @@ 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-0.2.51}/src/mcp_agent/llm/model_factory.py

@@ -22,7 +22,7 @@ from mcp_agent.llm.providers.augmented_llm_google_oai import GoogleOaiAugmentedL
 from mcp_agent.llm.providers.augmented_llm_groq import GroqAugmentedLLM
 from mcp_agent.llm.providers.augmented_llm_openai import OpenAIAugmentedLLM
 from mcp_agent.llm.providers.augmented_llm_openrouter import OpenRouterAugmentedLLM
-from mcp_agent.llm.providers.augmented_llm_tensorzero import TensorZeroAugmentedLLM
+from mcp_agent.llm.providers.augmented_llm_tensorzero_openai import TensorZeroOpenAIAugmentedLLM
 from mcp_agent.llm.providers.augmented_llm_xai import XAIAugmentedLLM
 from mcp_agent.mcp.interfaces import AugmentedLLMProtocol
 
@@ -39,7 +39,7 @@ LLMClass = Union[
     Type[SlowLLM],
     Type[DeepSeekAugmentedLLM],
     Type[OpenRouterAugmentedLLM],
-    Type[TensorZeroAugmentedLLM],
+    Type[TensorZeroOpenAIAugmentedLLM],
     Type[GoogleNativeAugmentedLLM],
     Type[GenericAugmentedLLM],
     Type[AzureOpenAIAugmentedLLM],
@@ -144,6 +144,8 @@ class ModelFactory:
         "gemini25": "gemini-2.5-flash-preview-05-20",
         "gemini25pro": "gemini-2.5-pro-preview-05-06",
         "kimi": "groq.moonshotai/kimi-k2-instruct",
+        "gpt-oss": "groq.openai/gpt-oss-120b",
+        "gpt-oss-20b": "groq.openai/gpt-oss-20b",
     }
 
     # Mapping of providers to their LLM classes
@@ -157,7 +159,7 @@ class ModelFactory:
         Provider.GOOGLE: GoogleNativeAugmentedLLM,
         Provider.XAI: XAIAugmentedLLM,
         Provider.OPENROUTER: OpenRouterAugmentedLLM,
-        Provider.TENSORZERO: TensorZeroAugmentedLLM,
+        Provider.TENSORZERO: TensorZeroOpenAIAugmentedLLM,
         Provider.AZURE: AzureOpenAIAugmentedLLM,
         Provider.ALIYUN: AliyunAugmentedLLM,
         Provider.BEDROCK: BedrockAugmentedLLM,

{fast_agent_mcp-0.2.49 → fast_agent_mcp-0.2.51}/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-0.2.51}/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-0.2.51/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-0.2.51}/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 +391,14 @@ 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 in (
+                "deepseek-r1-distill-llama-70b",
+                "openai/gpt-oss-120b",
+                "openai/gpt-oss-20b",
+            ):
+                message_dict.pop("reasoning", None)
+                message_dict.pop("channel", None)
+
             messages.append(message_dict)
 
             message_text = message.content
@@ -412,9 +420,8 @@ 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 +437,7 @@ 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 +446,14 @@ 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(