Merge pull request #60 from CrackingShells/dev

Fixing MCP tool lifecycle state management

Author: LittleCoinCoin

docs/CHANGELOG.md

@@ -0,0 +1,21 @@
+## [0.4.3-dev.2](https://github.com/CrackingShells/Hatchling/compare/v0.4.3-dev.1...v0.4.3-dev.2) (2025-08-26)
+
+
+### Bug Fixes
+
+* **mcp:** disabled tools being called ([1d24986](https://github.com/CrackingShells/Hatchling/commit/1d249869b9d37c3b387f3197d8fc313055ac952a))
+
+## [0.4.3-dev.1](https://github.com/CrackingShells/Hatchling/compare/v0.4.2...v0.4.3-dev.1) (2025-08-26)
+
+
+### Bug Fixes
+
+* **mcp:** tools status after server reconnection ([bca9cc6](https://github.com/CrackingShells/Hatchling/commit/bca9cc6ffbadd0d3ef691b60364ea285700476fa))
+
+
+### Documentation
+
+* **dev:** remove design reports ([2dfd324](https://github.com/CrackingShells/Hatchling/commit/2dfd324689062f9622b5c9468414133aa6493a88))
+* instructions for AMD GPUs ([e2499bd](https://github.com/CrackingShells/Hatchling/commit/e2499bdaa768408df7279ed673a96092d4424155))
+* instructions to run hatchling ([9d18ac5](https://github.com/CrackingShells/Hatchling/commit/9d18ac5a03ee56942db2cf5ecf014869ca7b9261))
+* moving `CONTRIBUTING.md` in the `/docs` ([998d1b2](https://github.com/CrackingShells/Hatchling/commit/998d1b2345ebda16f566c761a5d508f43f44d0e8))

docs/articles/users/tutorials/Installation/docker-ollama-setup.md

@@ -18,9 +18,9 @@ This document provides instructions on how to set up and run Ollama for deployin
 - On Windows, install Windows Subsystem for Linux (WSL). Latest version is v2: [Official Microsoft Documentation](https://learn.microsoft.com/en-us/windows/wsl/install)
 - GPU Support:
   - For MacOS users with Apple Silicon chips (typically M series), you can **follow the instructions for CPU and ignore the GPU-related sections**
-  - For Windows & Linux with dedicated GPUs, we strongly recommend enabling GPU support to increase LLM output speed. On the computer with the GPU, do:
+  - For Windows & Linux with dedicated GPUs, we strongly recommend enabling GPU support to increase LLM output speed. We will be using the official documentation for each GPU type:
     - NVIDIA GPUs: [NVIDIA Container Toolkit Installation](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html)
-    - AMD GPUs: Nothing, you can move on.
+    - AMD GPUs: [AMD ROCm Installation](https://rocm.docs.amd.com/projects/radeon/en/latest/docs/install/native_linux/install-radeon.html)
 
 ## Setup with Docker Desktop
 
@@ -33,10 +33,11 @@ This document provides instructions on how to set up and run Ollama for deployin
    - Either enable integration with your default WSL distro (arrow 4.1) OR select a specific one (arrow 4.2)
    - Click "Apply & Restart" if you make changes (arrow 5)
 
-3. **For NVIDIA GPU owners, setup GPU Support (nothing to do for AMD GPU owners at this stage)**:
+3. **For GPU owners, setup GPU Support**:
    - [Open a terminal](../../../appendices/open_a_terminal.md) on the computer with the GPU you want to use (for GPU servers, you likely connect through ssh)
      - On Windows, launch the Linux version that was installed via WSL and that Docker is using. For example, in the previous image, that would be `Ubuntu-24.04`; so, run `wsl -d Ubuntu-24.04` to start Ubuntu.
-   - For NVIDIA GPU support, run:
+   
+   - **For NVIDIA GPU support**, run:
 
      ```bash
      # Add NVIDIA repository keys
@@ -55,6 +56,24 @@ This document provides instructions on how to set up and run Ollama for deployin
      sudo nvidia-ctk runtime configure --runtime=docker
      ```
 
+   - **For AMD GPU support**, run:
+
+     ```bash
+     # Install required packages
+     sudo apt install python3-setuptools python3-wheel
+     
+     # Download and install AMD GPU installer script (for Ubuntu 24.04)
+     sudo apt update
+     wget https://repo.radeon.com/amdgpu-install/6.4.2.1/ubuntu/noble/amdgpu-install_6.4.60402-1_all.deb
+     sudo apt install ./amdgpu-install_6.4.60402-1_all.deb
+     
+     # Install graphics and ROCm support
+     sudo amdgpu-install -y --usecase=graphics,rocm
+     
+     # Add current user to render and video groups
+     sudo usermod -a -G render,video $LOGNAME
+     ```
+
    - Close the terminal
    - Restart Docker
      - For Docker Desktop, click on the three vertical dots icon (arrow 1), then `Restart` (arrow 2)

docs/articles/users/tutorials/Installation/running_hatchling.md

@@ -32,6 +32,9 @@ This section assumes you have followed the [Docker & Ollama setup](./docker-olla
   docker run -d --device /dev/kfd --device /dev/dri -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama:rocm
   ```
 
+  > [!Note]
+  > Troubleshooting: If you encounter issues with the `/dev/kfd` or `/dev/dri` devices, try running the command with the `--privileged` flag: `docker run -d --privileged --device /dev/kfd --device /dev/dri -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama:rocm`
+
 ### Checking that GPU support is enabled  as expected
 
 - Go to the `Containers` tab in Docker Desktop (arrow 1) and select your Ollama container
@@ -72,6 +75,12 @@ At this step, you will be downloading the content of Hatchling. Currently, we ar
 cd ./Hatchling/docker
 ```
 
+### Copy the `.env.example` file to `.env`
+
+```bash
+cp .env.example .env
+```
+
 ### Install Hatchling by building the code
 
    ```bash

hatchling/mcp_utils/mcp_tool_lifecycle_subscriber.py

@@ -121,39 +121,35 @@ def _handle_server_reachable_event(self, event: Event) -> None:
     def _handle_tool_enabled_event(self, event: Event) -> None:
         """Handle tool enabled event."""
         tool_name = event.data.get("tool_name", "")
-        
-        # Create or update tool info from event data
-        if tool_name not in self._tool_cache:
-            tool_info = event.data.get("tool_info", {})
+        tool_info = event.data.get("tool_info", {})
+
+        if not tool_info:
+            self.logger.error(f"'Tool enabled event' missing 'tool_info' for tool '{tool_name}'")
+            return
 
-            if not tool_info:
-                self.logger.error(f"'Tool enabled event' missing 'tool_info' for tool '{tool_name}'")
-                return
-            
-            # Convert tool to provider-specific format
-            # Tool info is an in/out parameter in mcp_to_provider_tool
-            # Hence, the provider_format field will be set
-            # to the converted tool format
-            self._mcp_to_provider_tool_func(tool_info)
+        # Always convert tool to provider-specific format
+        # This ensures that tools are properly formatted even during reconnection
+        # when they already exist in the cache but need their provider_format refreshed
+        self._mcp_to_provider_tool_func(tool_info)
 
-            self._tool_cache[tool_name] = tool_info
-            self.logger.debug(f"Tool enabled: {tool_name}")
+        # Update cache with the new tool info (whether new or existing)
+        self._tool_cache[tool_name] = tool_info
+        self.logger.debug(f"Tool enabled: {tool_name}")
 
     def _handle_tool_disabled_event(self, event: Event) -> None:
         """Handle tool disabled event."""
         tool_name = event.data.get("tool_name", "")
-        
-        if tool_name in self._tool_cache:
-            tool_info = self._tool_cache[tool_name]
-            tool_info.status = MCPToolStatus.DISABLED
-            
-            if "reason" in event.data:
-                try:
-                    tool_info.reason = MCPToolStatusReason[event.data["reason"].upper()]
-                except (KeyError, ValueError):
-                    tool_info.reason = MCPToolStatusReason.FROM_SYSTEM_ERROR
-            
-            self.logger.debug(f"Tool disabled: {tool_name}")
+        tool_info = event.data.get("tool_info", {})
+
+        if not tool_info:
+            self.logger.error(f"'Tool disabled event' missing 'tool_info' for tool '{tool_name}'")
+            return
+
+        # Update cache with the fresh tool info from the event
+        # This ensures consistency with _handle_tool_enabled_event() and prevents
+        # stale cache entries that could cause disabled tools to remain in payloads
+        self._tool_cache[tool_name] = tool_info
+        self.logger.debug(f"Tool disabled: {tool_name}")
 
     def get_enabled_tools(self) -> Dict[str, MCPToolInfo]:
         """Get all currently enabled tools.