diff --git a/.github/workflows/issue-triage-llm.lock.yml b/.github/workflows/issue-triage-llm.lock.yml
index 487d165020c..8330fc4e95a 100644
--- a/.github/workflows/issue-triage-llm.lock.yml
+++ b/.github/workflows/issue-triage-llm.lock.yml
@@ -1412,6 +1412,9 @@ jobs:
         run: |
           pip install llm llm-github-models llm-tools-mcp
           llm --version
+
+          # Show logs database path for debugging
+          echo "LLM logs database: $(llm logs path)"
         env:
           GITHUB_AW_MCP_CONFIG: /tmp/gh-aw/mcp-config/mcp-servers.json
           GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
@@ -1425,11 +1428,19 @@ jobs:
           if [ -n "$GITHUB_AW_MCP_CONFIG" ] && [ -f "$GITHUB_AW_MCP_CONFIG" ]; then
             # Create llm-tools-mcp config directory
             mkdir -p ~/.llm-tools-mcp
+            mkdir -p ~/.llm-tools-mcp/logs
 
             # Copy MCP configuration to the expected location for llm-tools-mcp
             cp "$GITHUB_AW_MCP_CONFIG" ~/.llm-tools-mcp/mcp.json
 
             echo "✓ MCP configuration installed at ~/.llm-tools-mcp/mcp.json"
+            echo "📋 MCP configuration:"
+            cat ~/.llm-tools-mcp/mcp.json
+            echo ""
+
+            # List available tools for debugging
+            echo "🔧 Listing available MCP tools:"
+            llm tools list || echo "⚠ Failed to list tools"
           else
             echo "ℹ No MCP configuration available"
           fi
@@ -1438,6 +1449,7 @@ jobs:
           GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
           GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          LLM_TOOLS_MCP_FULL_ERRORS: "1"
       - name: Run llm CLI with prompt
         id: llm_execution
         run: |
@@ -1450,12 +1462,15 @@ jobs:
 
           # Run llm with the prompt from the file
           # Use -T MCP to enable MCP tools if configured
+          # Additional flags for debugging:
+          #   --td: Show full details of tool executions
+          #   -u: Show token usage
           if [ -f ~/.llm-tools-mcp/mcp.json ]; then
-            echo "Running with MCP tools enabled"
-            llm -m "$MODEL" -T MCP "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
+            echo "🚀 Running with MCP tools enabled (debug mode)"
+            llm -m "$MODEL" -T MCP --td -u "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
           else
             echo "Running without MCP tools"
-            llm -m "$MODEL" "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
+            llm -m "$MODEL" -u "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
           fi
 
           # Store output for safe-outputs processing if configured
@@ -1467,6 +1482,18 @@ jobs:
           GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
           GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          LLM_TOOLS_MCP_FULL_ERRORS: "1"
+      - name: Upload MCP server logs
+        if: always()
+        uses: actions/upload-artifact@v4
+        env:
+          GITHUB_AW_MCP_CONFIG: /tmp/gh-aw/mcp-config/mcp-servers.json
+          GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
+          GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
+        with:
+          if-no-files-found: ignore
+          name: llm-mcp-logs
+          path: ~/.llm-tools-mcp/logs/
       - name: Ensure log file exists
         run: |
           echo "Custom steps execution completed" >> /tmp/gh-aw/agent-stdio.log
@@ -2285,7 +2312,7 @@ jobs:
           AGENT_OUTPUT: ${{ needs.agent.outputs.output }}
           WORKFLOW_NAME: "Issue Triage (LLM)"
           WORKFLOW_DESCRIPTION: "No description provided"
-          WORKFLOW_MARKDOWN: "<!--\nThis shared configuration sets up a custom agentic engine using simonw/llm CLI with GitHub Models.\n\n**Usage:**\nInclude this file in your workflow using frontmatter imports:\n\n```yaml\n---\nimports:\n  - shared/simonw-llm.md\n---\n```\n\n**Requirements:**\n- The workflow uses GitHub Models via the built-in GITHUB_TOKEN (no additional setup required)\n- The llm CLI will be installed via pip along with:\n  - llm-github-models: GitHub Models integration (free tier)\n  - llm-tools-mcp: MCP server support for tool access\n\n**Model:**\n- Uses `github/gpt-4o-mini` by default (free via GitHub Models)\n\n**MCP Tools:**\n- The llm-tools-mcp plugin enables MCP server integration\n- MCP configuration from GITHUB_AW_MCP_CONFIG is copied to `~/.llm-tools-mcp/mcp.json`\n- Tools from MCP servers are automatically enabled via the `-T MCP` flag when MCP config is present\n- MCP config file must be in the format expected by llm-tools-mcp (see https://github.com/VirtusLab/llm-tools-mcp)\n\n**Note**: \n- This workflow requires internet access to install Python packages\n- The llm CLI stores conversations in a local SQLite database\n- Output is automatically captured for safe-outputs processing\n- You can customize the model by modifying the MODEL variable in the run step\n- GitHub Models provides free access to 30+ AI models\n-->\n\n# Issue Triage\n\nYou are an issue triage assistant. Your task is to analyze newly created issues and provide helpful triage information.\n\n## Current Issue\n\n- **Issue Number**: ${{ github.event.inputs.issue_number || github.event.issue.number }}\n- **Repository**: ${{ github.repository }}\n- **Issue Content**: \n  ```\n  ${{ needs.activation.outputs.text }}\n  ```\n\n## Triage Guidelines\n\nPlease analyze the issue and provide:\n\n1. **Issue Type Classification**: Determine if this is a:\n   - Bug report (something broken or not working)\n   - Feature request (new functionality)\n   - Documentation update\n   - Question or support request\n   - Enhancement (improvement to existing feature)\n\n2. **Priority Assessment**: Based on the content, suggest a priority level:\n   - Critical (security, data loss, complete breakage)\n   - High (major functionality affected)\n   - Medium (important but not blocking)\n   - Low (minor issue or nice-to-have)\n\n3. **Initial Analysis**: Provide:\n   - A brief summary of the issue\n   - Any missing information that would be helpful\n   - Suggested next steps or questions to ask the reporter\n   - Related components or areas of the codebase that might be affected\n\n## Your Task\n\nAnalyze the issue content above and create a triage comment that includes:\n- The issue type classification\n- The suggested priority level\n- Your initial analysis and recommendations\n\nFormat your response as a helpful comment that will be added to the issue. Use clear formatting with sections and bullet points. Be professional and constructive.\n\n**Important**: Generate a single, well-formatted comment that provides value to both the issue reporter and repository maintainers.\n"
+          WORKFLOW_MARKDOWN: "<!--\nThis shared configuration sets up a custom agentic engine using simonw/llm CLI with GitHub Models.\n\n**Usage:**\nInclude this file in your workflow using frontmatter imports:\n\n```yaml\n---\nimports:\n  - shared/simonw-llm.md\n---\n```\n\n**Requirements:**\n- The workflow uses GitHub Models via the built-in GITHUB_TOKEN (no additional setup required)\n- The llm CLI will be installed via pip along with:\n  - llm-github-models: GitHub Models integration (free tier)\n  - llm-tools-mcp: MCP server support for tool access\n\n**Model:**\n- Uses `github/gpt-4o-mini` by default (free via GitHub Models)\n\n**MCP Tools:**\n- The llm-tools-mcp plugin enables MCP server integration\n- MCP configuration from GITHUB_AW_MCP_CONFIG is copied to `~/.llm-tools-mcp/mcp.json`\n- Tools from MCP servers are automatically enabled via the `-T MCP` flag when MCP config is present\n- MCP config file must be in the format expected by llm-tools-mcp (see https://github.com/VirtusLab/llm-tools-mcp)\n\n**Debugging and Logging:**\nThis configuration includes maximum debug tracing for troubleshooting MCP server issues:\n\n1. **Environment Variables:**\n   - `LLM_TOOLS_MCP_FULL_ERRORS=1` - Enables full error stack traces for MCP connection failures\n   - Set on both \"Configure llm\" and \"Run llm CLI\" steps\n\n2. **CLI Flags:**\n   - `--td` (--tools-debug) - Shows full details of tool executions\n   - `-u` (--usage) - Shows token usage information\n\n3. **MCP Server Logs:**\n   - MCP server connection logs are written to `~/.llm-tools-mcp/logs/`\n   - These logs are uploaded as workflow artifacts (artifact name: `llm-mcp-logs`)\n   - Each MCP server session creates a timestamped log file\n\n4. **Diagnostic Output:**\n   - MCP configuration is printed to stdout during setup\n   - Available MCP tools are listed with `llm tools list`\n   - LLM logs database path is displayed\n\n5. **Log Locations:**\n   - MCP server logs: `~/.llm-tools-mcp/logs/` (uploaded as artifacts)\n   - LLM conversation logs: View with `llm logs path` command\n   - Workflow output: `/tmp/gh-aw/llm-output.txt`\n\n**Troubleshooting MCP Issues:**\nIf MCP servers fail to load:\n1. Check the workflow run artifacts for `llm-mcp-logs`\n2. Review the \"Configure llm with GitHub Models and MCP\" step output for configuration details\n3. Check the \"Run llm CLI\" step output for tool execution details\n4. Look for error messages with full stack traces (enabled by LLM_TOOLS_MCP_FULL_ERRORS)\n\n**Note**: \n- This workflow requires internet access to install Python packages\n- The llm CLI stores conversations in a local SQLite database\n- Output is automatically captured for safe-outputs processing\n- You can customize the model by modifying the MODEL variable in the run step\n- GitHub Models provides free access to 30+ AI models\n-->\n\n# Issue Triage\n\nYou are an issue triage assistant. Your task is to analyze newly created issues and provide helpful triage information.\n\n## Current Issue\n\n- **Issue Number**: ${{ github.event.inputs.issue_number || github.event.issue.number }}\n- **Repository**: ${{ github.repository }}\n- **Issue Content**: \n  ```\n  ${{ needs.activation.outputs.text }}\n  ```\n\n## Triage Guidelines\n\nPlease analyze the issue and provide:\n\n1. **Issue Type Classification**: Determine if this is a:\n   - Bug report (something broken or not working)\n   - Feature request (new functionality)\n   - Documentation update\n   - Question or support request\n   - Enhancement (improvement to existing feature)\n\n2. **Priority Assessment**: Based on the content, suggest a priority level:\n   - Critical (security, data loss, complete breakage)\n   - High (major functionality affected)\n   - Medium (important but not blocking)\n   - Low (minor issue or nice-to-have)\n\n3. **Initial Analysis**: Provide:\n   - A brief summary of the issue\n   - Any missing information that would be helpful\n   - Suggested next steps or questions to ask the reporter\n   - Related components or areas of the codebase that might be affected\n\n## Your Task\n\nAnalyze the issue content above and create a triage comment that includes:\n- The issue type classification\n- The suggested priority level\n- Your initial analysis and recommendations\n\nFormat your response as a helpful comment that will be added to the issue. Use clear formatting with sections and bullet points. Be professional and constructive.\n\n**Important**: Generate a single, well-formatted comment that provides value to both the issue reporter and repository maintainers.\n"
         with:
           script: |
             const fs = require('fs');
@@ -2370,6 +2397,9 @@ jobs:
         run: |
           pip install llm llm-github-models llm-tools-mcp
           llm --version
+
+          # Show logs database path for debugging
+          echo "LLM logs database: $(llm logs path)"
         env:
           GITHUB_AW_MCP_CONFIG: /tmp/gh-aw/mcp-config/mcp-servers.json
           GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
@@ -2383,11 +2413,19 @@ jobs:
           if [ -n "$GITHUB_AW_MCP_CONFIG" ] && [ -f "$GITHUB_AW_MCP_CONFIG" ]; then
             # Create llm-tools-mcp config directory
             mkdir -p ~/.llm-tools-mcp
+            mkdir -p ~/.llm-tools-mcp/logs
 
             # Copy MCP configuration to the expected location for llm-tools-mcp
             cp "$GITHUB_AW_MCP_CONFIG" ~/.llm-tools-mcp/mcp.json
 
             echo "✓ MCP configuration installed at ~/.llm-tools-mcp/mcp.json"
+            echo "📋 MCP configuration:"
+            cat ~/.llm-tools-mcp/mcp.json
+            echo ""
+
+            # List available tools for debugging
+            echo "🔧 Listing available MCP tools:"
+            llm tools list || echo "⚠ Failed to list tools"
           else
             echo "ℹ No MCP configuration available"
           fi
@@ -2396,6 +2434,7 @@ jobs:
           GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
           GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          LLM_TOOLS_MCP_FULL_ERRORS: "1"
       - name: Run llm CLI with prompt
         id: llm_execution
         run: |
@@ -2408,12 +2447,15 @@ jobs:
 
           # Run llm with the prompt from the file
           # Use -T MCP to enable MCP tools if configured
+          # Additional flags for debugging:
+          #   --td: Show full details of tool executions
+          #   -u: Show token usage
           if [ -f ~/.llm-tools-mcp/mcp.json ]; then
-            echo "Running with MCP tools enabled"
-            llm -m "$MODEL" -T MCP "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
+            echo "🚀 Running with MCP tools enabled (debug mode)"
+            llm -m "$MODEL" -T MCP --td -u "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
           else
             echo "Running without MCP tools"
-            llm -m "$MODEL" "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
+            llm -m "$MODEL" -u "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
           fi
 
           # Store output for safe-outputs processing if configured
@@ -2425,6 +2467,17 @@ jobs:
           GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
           GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          LLM_TOOLS_MCP_FULL_ERRORS: "1"
+      - name: Upload MCP server logs
+        if: always()
+        uses: actions/upload-artifact@v4
+        env:
+          GITHUB_AW_MCP_CONFIG: /tmp/gh-aw/mcp-config/mcp-servers.json
+          GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
+        with:
+          if-no-files-found: ignore
+          name: llm-mcp-logs
+          path: ~/.llm-tools-mcp/logs/
       - name: Ensure log file exists
         run: |
           echo "Custom steps execution completed" >> /tmp/gh-aw/threat-detection/detection.log
diff --git a/.github/workflows/shared/simonw-llm.lock.yml b/.github/workflows/shared/simonw-llm.lock.yml
index fe58c947b6a..0dd0dd5460f 100644
--- a/.github/workflows/shared/simonw-llm.lock.yml
+++ b/.github/workflows/shared/simonw-llm.lock.yml
@@ -211,7 +211,7 @@ jobs:
                   "--rm",
                   "-e",
                   "GITHUB_PERSONAL_ACCESS_TOKEN",
-                  "ghcr.io/github/github-mcp-server:sha-09deac4"
+                  "ghcr.io/github/github-mcp-server:v0.18.0"
                 ],
                 "env": {
                   "GITHUB_PERSONAL_ACCESS_TOKEN": "${{ secrets.GH_AW_GITHUB_TOKEN || secrets.GITHUB_TOKEN }}"
@@ -344,9 +344,13 @@ jobs:
         run: |
           pip install llm llm-github-models llm-tools-mcp
           llm --version
+
+          # Show logs database path for debugging
+          echo "LLM logs database: $(llm logs path)"
         env:
           GITHUB_AW_MCP_CONFIG: /tmp/gh-aw/mcp-config/mcp-servers.json
           GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
+          GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
       - name: Configure llm with GitHub Models and MCP
         run: |
           # GitHub Models uses GITHUB_TOKEN by default, no key setup needed
@@ -356,18 +360,28 @@ jobs:
           if [ -n "$GITHUB_AW_MCP_CONFIG" ] && [ -f "$GITHUB_AW_MCP_CONFIG" ]; then
             # Create llm-tools-mcp config directory
             mkdir -p ~/.llm-tools-mcp
+            mkdir -p ~/.llm-tools-mcp/logs
 
             # Copy MCP configuration to the expected location for llm-tools-mcp
             cp "$GITHUB_AW_MCP_CONFIG" ~/.llm-tools-mcp/mcp.json
 
             echo "✓ MCP configuration installed at ~/.llm-tools-mcp/mcp.json"
+            echo "📋 MCP configuration:"
+            cat ~/.llm-tools-mcp/mcp.json
+            echo ""
+
+            # List available tools for debugging
+            echo "🔧 Listing available MCP tools:"
+            llm tools list || echo "⚠ Failed to list tools"
           else
             echo "ℹ No MCP configuration available"
           fi
         env:
           GITHUB_AW_MCP_CONFIG: /tmp/gh-aw/mcp-config/mcp-servers.json
           GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
+          GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          LLM_TOOLS_MCP_FULL_ERRORS: "1"
       - name: Run llm CLI with prompt
         id: llm_execution
         run: |
@@ -380,12 +394,15 @@ jobs:
 
           # Run llm with the prompt from the file
           # Use -T MCP to enable MCP tools if configured
+          # Additional flags for debugging:
+          #   --td: Show full details of tool executions
+          #   -u: Show token usage
           if [ -f ~/.llm-tools-mcp/mcp.json ]; then
-            echo "Running with MCP tools enabled"
-            llm -m "$MODEL" -T MCP "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
+            echo "🚀 Running with MCP tools enabled (debug mode)"
+            llm -m "$MODEL" -T MCP --td -u "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
           else
             echo "Running without MCP tools"
-            llm -m "$MODEL" "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
+            llm -m "$MODEL" -u "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
           fi
 
           # Store output for safe-outputs processing if configured
@@ -397,6 +414,17 @@ jobs:
           GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
           GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          LLM_TOOLS_MCP_FULL_ERRORS: "1"
+      - name: Upload MCP server logs
+        if: always()
+        uses: actions/upload-artifact@v4
+        env:
+          GITHUB_AW_MCP_CONFIG: /tmp/gh-aw/mcp-config/mcp-servers.json
+          GITHUB_AW_PROMPT: /tmp/gh-aw/aw-prompts/prompt.txt
+        with:
+          if-no-files-found: ignore
+          name: llm-mcp-logs
+          path: ~/.llm-tools-mcp/logs/
       - name: Ensure log file exists
         run: |
           echo "Custom steps execution completed" >> /tmp/gh-aw/agent-stdio.log
diff --git a/.github/workflows/shared/simonw-llm.md b/.github/workflows/shared/simonw-llm.md
index 6c4bf3cd9bc..50f7fd72c6f 100644
--- a/.github/workflows/shared/simonw-llm.md
+++ b/.github/workflows/shared/simonw-llm.md
@@ -6,8 +6,13 @@ engine:
       run: |
         pip install llm llm-github-models llm-tools-mcp
         llm --version
+        
+        # Show logs database path for debugging
+        echo "LLM logs database: $(llm logs path)"
       env:
         GITHUB_AW_MCP_CONFIG: ${{ env.GITHUB_AW_MCP_CONFIG }}
+        GITHUB_AW_PROMPT: ${{ env.GITHUB_AW_PROMPT }}
+        GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
     
     - name: Configure llm with GitHub Models and MCP
       run: |
@@ -18,17 +23,28 @@ engine:
         if [ -n "$GITHUB_AW_MCP_CONFIG" ] && [ -f "$GITHUB_AW_MCP_CONFIG" ]; then
           # Create llm-tools-mcp config directory
           mkdir -p ~/.llm-tools-mcp
+          mkdir -p ~/.llm-tools-mcp/logs
           
           # Copy MCP configuration to the expected location for llm-tools-mcp
           cp "$GITHUB_AW_MCP_CONFIG" ~/.llm-tools-mcp/mcp.json
           
           echo "✓ MCP configuration installed at ~/.llm-tools-mcp/mcp.json"
+          echo "📋 MCP configuration:"
+          cat ~/.llm-tools-mcp/mcp.json
+          echo ""
+          
+          # List available tools for debugging
+          echo "🔧 Listing available MCP tools:"
+          llm tools list || echo "⚠ Failed to list tools"
         else
           echo "ℹ No MCP configuration available"
         fi
       env:
-        GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         GITHUB_AW_MCP_CONFIG: ${{ env.GITHUB_AW_MCP_CONFIG }}
+        GITHUB_AW_PROMPT: ${{ env.GITHUB_AW_PROMPT }}
+        GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
+        GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        LLM_TOOLS_MCP_FULL_ERRORS: "1"
     
     - name: Run llm CLI with prompt
       id: llm_execution
@@ -42,12 +58,15 @@ engine:
         
         # Run llm with the prompt from the file
         # Use -T MCP to enable MCP tools if configured
+        # Additional flags for debugging:
+        #   --td: Show full details of tool executions
+        #   -u: Show token usage
         if [ -f ~/.llm-tools-mcp/mcp.json ]; then
-          echo "Running with MCP tools enabled"
-          llm -m "$MODEL" -T MCP "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
+          echo "🚀 Running with MCP tools enabled (debug mode)"
+          llm -m "$MODEL" -T MCP --td -u "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
         else
           echo "Running without MCP tools"
-          llm -m "$MODEL" "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
+          llm -m "$MODEL" -u "$(cat $GITHUB_AW_PROMPT)" 2>&1 | tee /tmp/gh-aw/llm-output.txt
         fi
         
         # Store output for safe-outputs processing if configured
@@ -59,6 +78,15 @@ engine:
         GITHUB_AW_SAFE_OUTPUTS: ${{ env.GITHUB_AW_SAFE_OUTPUTS }}
         GITHUB_AW_MCP_CONFIG: ${{ env.GITHUB_AW_MCP_CONFIG }}
         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        LLM_TOOLS_MCP_FULL_ERRORS: "1"
+    
+    - name: Upload MCP server logs
+      if: always()
+      uses: actions/upload-artifact@v4
+      with:
+        name: llm-mcp-logs
+        path: ~/.llm-tools-mcp/logs/
+        if-no-files-found: ignore
 ---
 
 <!--
@@ -89,6 +117,39 @@ imports:
 - Tools from MCP servers are automatically enabled via the `-T MCP` flag when MCP config is present
 - MCP config file must be in the format expected by llm-tools-mcp (see https://github.com/VirtusLab/llm-tools-mcp)
 
+**Debugging and Logging:**
+This configuration includes maximum debug tracing for troubleshooting MCP server issues:
+
+1. **Environment Variables:**
+   - `LLM_TOOLS_MCP_FULL_ERRORS=1` - Enables full error stack traces for MCP connection failures
+   - Set on both "Configure llm" and "Run llm CLI" steps
+
+2. **CLI Flags:**
+   - `--td` (--tools-debug) - Shows full details of tool executions
+   - `-u` (--usage) - Shows token usage information
+
+3. **MCP Server Logs:**
+   - MCP server connection logs are written to `~/.llm-tools-mcp/logs/`
+   - These logs are uploaded as workflow artifacts (artifact name: `llm-mcp-logs`)
+   - Each MCP server session creates a timestamped log file
+
+4. **Diagnostic Output:**
+   - MCP configuration is printed to stdout during setup
+   - Available MCP tools are listed with `llm tools list`
+   - LLM logs database path is displayed
+
+5. **Log Locations:**
+   - MCP server logs: `~/.llm-tools-mcp/logs/` (uploaded as artifacts)
+   - LLM conversation logs: View with `llm logs path` command
+   - Workflow output: `/tmp/gh-aw/llm-output.txt`
+
+**Troubleshooting MCP Issues:**
+If MCP servers fail to load:
+1. Check the workflow run artifacts for `llm-mcp-logs`
+2. Review the "Configure llm with GitHub Models and MCP" step output for configuration details
+3. Check the "Run llm CLI" step output for tool execution details
+4. Look for error messages with full stack traces (enabled by LLM_TOOLS_MCP_FULL_ERRORS)
+
 **Note**: 
 - This workflow requires internet access to install Python packages
 - The llm CLI stores conversations in a local SQLite database