AI Personal Hub

Unified personal assistant that connects Local Files, GitHub, YouTube, Gmail, Steam, and Summarization through Model Context Protocol (MCP), with a clean Flask UI and LM Studio integration.

Features

• Local notes: list and open .txt files from notes/
• GitHub: repos, commits, list files, fetch file content, issues
• YouTube: Liked Videos (LL), Liked Songs (LM) via OAuth
• Gmail: read last emails via OAuth
• Steam: recent owned games and playtime
• Summarizer: prompt exposed as an MCP tool
• Modern chat-style UI with quick actions and linkified results

Architecture

• Flask serves the UI and a simple /ask endpoint
• LM Studio runs a local OpenAI-compatible server for LLM responses
• MCP server (mcp_server/server.py) exposes tools that LM Studio can call

Browser ↔ Flask UI ↔ LM Studio (LLM) ↔ MCP Tools (python -m mcp_server.server)

Setup

1. Python env

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt

2. LM Studio (recommended defaults)

• Start LM Studio local server (OpenAI-compatible) on http://localhost:1234
• Set environment variables (PowerShell):

setx LM_STUDIO_BASE_URL http://localhost:1234
setx LM_STUDIO_API_KEY lm-studio
setx LM_STUDIO_MODEL your-model-name

3. MCP server registration in LM Studio

Edit LM Studio mcp.json and add:

{
  "mcpServers": {
    "personal-hub-server": {
      "command": "D:\\AI_MCP\\ai-personal-hub\\.venv\\Scripts\\python.exe",
      "args": ["-m", "mcp_server.server"],
      "cwd": "D:\\AI_MCP\\ai-personal-hub"
    }
  }
}

Then start the server in LM Studio Tools (MCP), or run manually:

python -m mcp_server.server

4. Service credentials

• GitHub: set GITHUB_TOKEN (PAT, repo read scope recommended)
• Steam: set STEAM_API_KEY and STEAM_ID
• Notes: create notes/ with .txt files

YouTube OAuth (token.json)

• In Google Cloud: enable "YouTube Data API v3"
• Create OAuth client (Desktop app) → download client_secret.json to project root
• Generate token.json (one-time):

python -c "from google_auth_oauthlib.flow import InstalledAppFlow; import json; flow=InstalledAppFlow.from_client_secrets_file('client_secret.json',['https://www.googleapis.com/auth/youtube.readonly']); creds=flow.run_local_server(port=0); open('token.json','w',encoding='utf-8').write(creds.to_json())"

Gmail OAuth (token.json)

• Similar flow; ensure scope https://www.googleapis.com/auth/gmail.readonly

Run

Flask UI:

python app.py

Open http://127.0.0.1:5000/

MCP server (separate terminal):

python -m mcp_server.server

Available MCP tools

• Files: list_local_files, fetch_local_file
• GitHub: github_repos, github_commits, github_list_files, github_file_content, github_issues, github_issue
• YouTube: yt_liked_videos, ytm_liked_songs, yt_playlist
• Gmail: read_emails
• Steam: steam_games
• Summarize: summarize prompt

Example prompts

• “List my local notes.”
• “Open a.txt.”
• “List repos for Harsh-1807.”
• “List files in Harsh-1807/weather.”
• “Open README.md from Harsh-1807/weather.”
• “Show open issues for Harsh-1807/weather.”
• “List 5 of my liked YouTube videos.”
• “List 5 of my liked songs on YouTube Music.”
• “Summarize my last 5 emails.”
• “Which Steam games do I play most?”

Troubleshooting

• LLM answers without calling tools: lower temperature; add a system prompt telling it to prefer MCP tools; ensure the tool server is running and registered in LM Studio.
• YouTube LM liked songs not returning results: some accounts do not expose LM over the API; try yt_liked_videos (LL) or yt_playlist("LL").
• Permission errors: re-create token.json for the correct Google account; verify scopes.

License

MIT