# Giordano Desktop Fix — Next Session **Date:** 2026-08-03 **Machine:** Windows, AMD Ryzen 3 2200G, 8GB RAM, user `giord`, Python 3.14 **Issue:** Claude Desktop MCP connector fails — wrong MCP package version installed ## Prerequisites - Diligence app folder at `C:\Users\giord\Downloads\Diligence-main` - Claude Desktop installed (Microsoft Store version) - Node.js installed - `mcp_stdio.py` already in the Diligence-main folder - Claude Desktop config already set at both paths: - `%APPDATA%\Claude\claude_desktop_config.json` - `%LOCALAPPDATA%\Packages\Claude_pzs8sxrjxfjjc\LocalCache\Roaming\Claude\claude_desktop_config.json` ## Step 1 — Clean up old MCP packages ```powershell cd "$env:USERPROFILE\Downloads\Diligence-main" pip uninstall mcp mcp-cli fastmcp -y ``` ## Step 2 — Fix the import in server.py The old code imports from the wrong package. Run this once: ```powershell (Get-Content diligence\mcp\server.py) -replace 'from mcp\.server\.fastmcp import FastMCP', 'from fastmcp import FastMCP' | Set-Content diligence\mcp\server.py ``` ## Step 3 — Install the correct packages ```powershell pip install "fastmcp>=3.0.0" pip install tzdata ``` ## Step 4 — Test the MCP stdio script ```powershell python mcp_stdio.py ``` - **If it hangs with no output** → success. Ctrl+C to stop. - **If it throws an error** → paste the traceback. ## Step 5 — Start Diligence ```powershell python -m diligence ``` Confirm **both** lines appear: - `Diligence running at http://localhost:8000` - `MCP: http://localhost:3001/sse` ## Step 6 — Restart Claude Desktop ```powershell Get-Process *claude* | Stop-Process -Force ``` Reopen Claude Desktop from the Start menu. Look for a tools/hammer icon indicating MCP connected. ## Step 7 — Test In Claude Desktop, type: > What's my fitness status today? Claude should call the `get_context` tool and return fitness data. Also test a write operation: > Log a 30 minute workout called Morning Run --- ## Fallback: Built-in AI Coach (no Claude Desktop needed) If Claude Desktop MCP continues to have issues, use the built-in coach: 1. Go to `http://localhost:8000/agent` in the browser 2. Under "Option 1 — Built-in AI Coach", click **Get key** next to **OpenRouter** 3. Sign up with Google (free, 26 free models available) 4. Copy the API key 5. In Diligence, go to **Settings > Integrations > AI Coaching** 6. Paste the key and select a model 7. Use the **Coach** tab in the bottom nav This runs entirely inside the app — no external tools needed. --- ## Known Issues on This Machine | Issue | Fix | Status | |-------|-----|--------| | tzdata not in pip dependencies | pip install tzdata manually | Pending GitHub push | | mcp v2.0.0 removed FastMCP import | Switch to standalone fastmcp package | Fixed in subo repo, pending GitHub | | GitHub push token expired | Need new PAT with repo scope | Not yet done | | Migration warning: NOT NULL on point_rules.description | Non-fatal, cosmetic | Low priority | | PowerShell QuickEdit mode pauses the app | Right-click title bar, Properties, uncheck QuickEdit | Told user |