MemPlato
About
Personal MCP server that runs on your Android phone via Termux. 29 tools: drawers, knowledge graph, diary, semantic search, tunnels. All data stored locally on your device, never on external servers. Works with Claude, Perplexity, Cursor, Windsurf, VS Code, Zed.
Details
- Author
- VChe-creator
- Downloads
- 322
- Categories
- AI, Other, Knowledge Base
Jump to
- 29 tools including drawers, knowledge graph, diary, semantic search, tunnels
- All data stored locally on your device, never on external servers
- Runs on Android via Termux
- Works with Claude, Perplexity, Cursor, Windsurf, VS Code, Zed
Run MemPlato on your Android phone inside Termux, then point any supported MCP client (such as Claude, Perplexity, Cursor, Windsurf, VS Code, or Zed) to the server. No installation steps or configuration keys are listed beyond running it in Termux.
Personal MCP memory server that runs on your Android phone.
Every AI conversation starts from zero. MemPlato fixes that — it's a persistent memory layer for AI assistants, running entirely on your own device. Your data never leaves your phone.
MemPlato turns your Android phone into a private MCP (Model Context Protocol) server. Claude, Perplexity, Cursor, and other AI tools can connect to it and remember things across conversations — your projects, preferences, notes, and context — without sending any of it to a cloud service.
The phone stays in your pocket. The memory stays on your device.
Your Phone (Termux) Relay Server AI Assistant ───────────────── ───────────── ────────────── memplato_server.py ←──SSH── relay.memplato.com ←──MCP── Claude port 7333 (just a bridge) SQLite + ONNX your data never stored
👆 Click to watch the full installation walkthrough on YouTube
MemPlato runs a Python FastAPI server insideTermuxon your Android phone. A reverse SSH tunnel throughrelay.memplato.comgives AI clients a stable HTTPS endpoint to connect to. The relay is a dumb pipe — it never stores your data.
Tech stack:Flutter (Android app) · Python 3.13 · FastAPI · SQLite · ONNX Runtime · MCP SSE · autossh
- 29 MCP tools— drawers, knowledge graph, diary, tunnels, semantic search
- Semantic searchwith local ONNX embeddings (all-MiniLM-L6-v2, runs offline)
- Knowledge graphwith temporal facts (valid_from / valid_to)
- Cross-wing tunnels— link related memory clusters
- Agent diary— per-agent journal entries
- Your data stays on your phone— 100% of your memories are stored locally on your device, never on external servers
- Auto-reconnect watchdog— tunnel restarts automatically if it drops
- English + Ukrainian UI
The APK is ~280 MB because it includes Python, all libraries, and the AI model — so setup on your phone requires no internet connection.
- Android phone (Android 12+ recommended, ARM64)
- ~1 GB free storage
- Patience ☕
- InstallTermux from GitHub(not from Play Store — it's outdated)
- Install theMemPlato APK
- Grant permission in Android Settings
- Follow the in-app setup steps — the app guides you through everything
- TapInstall and Start— setup takes 30–90 minutes, keep screen on and charger plugged in
- When done, a green dot appears and you get your personal URL:
https://relay.memplato.com/u/YOUR_ID/mcp
- In Claude → Settings → Connectors → Add MCP server → paste your URL
Termux must stay running in the background (you'll see it in your notification bar). The watchdog script automatically restarts the tunnel if it drops. You don't need to do anything — just keep Termux open.
⚠️Known limitation:After rebooting your phone, the server does not restart automatically. Use theStartbutton in the app to restart it manually. Auto-start on boot will be added in a future update.
MemPlato v1.0.5 was tested with140+ test casescovering all 29 tools including edge cases, stress tests, and failure scenarios.
The server is stable and handles all edge cases without crashing. There are known limitations that will be addressed in the next major version:
- Embedding quality— The lightweight ONNX model (all-MiniLM-L6-v2) optimized for mobile occasionally produces unexpected similarity scores for unrelated texts. Search works reliably for standard use cases. Will be improved in v2 with a better model.
- Cross-lingual search— Ukrainian/English mixed queries sometimes return suboptimal ranking. For best results, use the same language as your stored content.
- Duplicate protection—add_drawerdoes not automatically callcheck_duplicate. Best practice: callcheck_duplicatemanually before saving.
- Search limit— Semantic search is optimized for queries returning up to 10 results. Requesting more may cause an error in some configurations.
- Auto-start after reboot— Auto-start after reboot — After rebooting your phone, use the Start button in the app to restart the server manually. Auto-start on boot will be added in a future update.
All known issues are cosmetic or edge-case. The core functionality — storing memories, knowledge graph, and MCP connectivity — works reliably.
Android Phone ├── MemPlato.apk (Flutter) │ ├── Manages setup flow │ ├── Copies all files to Termux │ └── Shows server status │ └── Termux (Linux environment) ├── memplato_server.py (FastAPI + MCP) │ ├── /sse ← MCP SSE endpoint │ ├── /mcp ← MCP HTTP endpoint │ └── /health ├── palace.db (SQLite) ├── models/onnx/ (384-dim embeddings) ├── autossh (reverse tunnel) └── tunnel_watchdog.sh (auto-restart) relay.memplato.com (Vultr VPS) └── nginx + relay.py └── Bridges AI clients to your phone └── No data stored
MemPlato requires Termux, which needs sideloading due to Android restrictions on terminal apps. The Google Play version of Termux is outdated and incompatible. This is a known limitation of the current MVP architecture.
The plan:After funding, MemPlato v2 will run without Termux — the server will be embedded directly into the app. No terminal setup, no manual steps. Straight to Google Play.
- MVP — Flutter app + Python server + MCP SSE
- All packages bundled in APK — no internet needed for setup
- Reverse SSH tunnel via relay
- 29 MCP tools
- English + Ukrainian localization
- v2 — server embedded in app, no Termux needed
- v2 — sync across multiple phones, computers and clouds
- Google Play release
- Better embedding model
- Manual server start/stop from the app
- Auto-start on phone boot
MemPlato is built by a solo founder. If you find it useful, you can support the project:
- ⭐Star this repo— it helps more than you think
- 🐛Report bugs— open anIssue
- 💬Share feedback— what features would you use?
- 💰Sponsor—GitHub Sponsors(coming soon)
Looking for pre-seed investment ($75K).Building toward Google Play launch and 10,000 users. Reach out:ceo@memplato.comormemplato.com
AGPL-3.0— free to use and modify, but derivative works must remain open source.
Authenticated MCP and agent gateway for Forge Cascade private AI memory, provenance, graph search, and capsule lineage.
Perseus Vault is a single Rust binary that gives AI agents durable memory across sessions. One binary. One file. No Docker. No Postgres. No cloud. Just persistent memory that works with any MCP host.
Self-hosted MCP server for persistent agent memory. Define typed schemas, write records, and query with SQL via DuckDB against Parquet files on disk.
Fulcra Context MCP server for accessing your personal health, workouts, sleep, location, and more, all privately. Built around Context by Fulcra.
AI trading memory layer for MT5/forex with 15 MCP tools — store/recall trades, pattern discovery, strategy evolution, and Outcome-Weighted Memory.
Persistent cloud memory for AI agents via MCP Streamable HTTP. Store, retrieve, and search key-value memories across sessions and devices. Trial account auto-created on first use.
real world, traceable spatial memory for fact verification about the world.
Privacy-first GPS tracking companion. Query travel statistics, manage waypoints, track countries visited, and analyze distances and speeds through natural language. Free, GDPR compliant.
MCP-native Web analytics. Install one JS snippet, add the connector to Claude/Cursor/ChatGPT, then ask questions about your traffic directly in chat. No dashboard. No charts. Just answers. EU-hosted, GDPR-ready. Free up to 100k hits/month.
Local-first RLHF feedback loop for AI agents — capture preference signals, promote memories, block repeated mistakes, export DPO/KTO training pairs
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.





