Open SourceMemVault
    AI persistent memory layer (MCP Server)Node.js (ESM) + Express + SQLite + MCP Protocol + Gemini AI Engine

    MemVault.

    Self-hosted MCP server that gives Claude, Cursor, Antigravity, and every AI tool persistent memory about you — your code, projects, habits, and preferences. Silently captures your digital footprint via 7 data engines (Git commits, VS Code projects, clipboard, file changes, system info, browser history, AI conversations) and 4 AI importers (ChatGPT, Claude, Gemini, Perplexity). Serves 20 MCP tools + 5 resources + 3 prompts to any connected AI client.

    View on GitHubInstallation Guide

    What the README tells you

    4 ⭐ | 0 forks | 12 commits | MIT licensed | Live demo at memvault-three.vercel.app. The README documents 20 MCP tools by phase (Core → Smart Context → AI Intelligence), 7 sync engines, 4 AI importers, project structure tree, MCP config examples for Claude Desktop/Cursor/VS Code/Antigravity, CLI commands, and a 7-phase roadmap.

    Features

    What this repo gives you.

    A quick practical feature map before you open the source code.

    20 MCP Tools — vault_smart_search (intent-driven semantic search), vault_smart_context (relevance-ranked results), vault_capture_prompt (auto-logs prompts), vault_project_context (full project intelligence), vault_ai_summarize, vault_ai_insights, vault_weekly_digest, vault_daily_digest, vault_remember, vault_git_log, vault_recent_files, vault_system_info, vault_projects, vault_search, vault_add, vault_list, vault_get_context, vault_stats, vault_secret_list, vault_log_conversation.

    7 Data Capture Engines — sync-git.mjs (Git commit history from all repos), sync-vscode.mjs (VS Code projects/extensions/settings), sync-clipboard.mjs (clipboard with classification, opt-in), sync-files.mjs (file change tracking), sync-system.mjs (OS/RAM/CPU/dev tools snapshot), sync-browser.mjs (Chrome/Edge history), sync-antigravity.mjs (AI conversation capture).

    4 AI Importers — import-chatgpt.mjs, import-claude.mjs, import-gemini.mjs, import-perplexity.mjs for one-command chat history import from data exports.

    Gemini AI Intelligence Layer — vault_ai_summarize, vault_ai_insights, vault_smart_search (semantic), vault_weekly_digest using Gemini 2.0 Flash API for auto-summarization and insight extraction.

    Smart Context Engine — Auto-tagging (react, python, trading, bugfix), relevance scoring, deduplication, memory recall, and daily/weekly digests.

    Web UI Dashboard — Dark-mode UI on localhost:7799 with diary entry writer, full-text search, vault statistics, encrypted secrets manager (AES-256-GCM), and access map visualization.

    Security — Local-first, zero-cloud: stdio MCP transport (never over network), AES-256-GCM encrypted secrets, no telemetry, clipboard opt-in by default.

    Installation

    Step-by-Step Setup.

    Follow these steps to get MemVault running on your machine.

    Setup Path

    Run it locally, then inspect the system.

    Follow the README flow first. Once the app opens locally, use the AI prompts below to trace the data flow and make focused changes.

    CloneInstallRunInspect
    1

    Install

    Start here

    Install Node.js v18+ from nodejs.org.

    2

    Clone

    Clone. git clone https://github.com/MrChartist/memvault.git && cd memvault

    git clone https://github.com/MrChartist/memvault.git
    3

    Install

    Install. npm install

    npm install
    4

    Step 4

    Run setup wizard. node init.mjs — configures vault data path, capture engines, and shows MCP config for your AI clients.

    node init.mjs
    5

    Start server

    Start server. npm start — web UI + API runs on http://localhost:7799.

    npm start
    6

    Step 6

    Sync data. npm run sync:all (or individually: npm run sync:git, sync:vscode, sync:system, sync:files).

    7

    Step 7

    Connect to AI clients: Add MCP config to Claude Desktop (claude_desktop_config.json), Cursor (Settings → MCP), VS Code (Cline/Roo), or Antigravity (~/.gemini/antigravity/mcp_config.json) — see README for exact JSON snippets.

    8

    Step 8

    Production

    (Optional) Configure Gemini AI: Set ai.apiKey in ~/.memvaultrc.json for summarization and insights.

    Want AI to do this for you?Copy a ready-made prompt for Antigravity, Codex, or Claude below ↓
    AI-Powered Setup

    Set Up MemVault With AI.

    Pick your AI tool, copy the prompt, and let it handle the entire setup and codebase walkthrough for you.

    Google Antigravity

    Best when you want the assistant to operate inside the editor and handle setup end-to-end.

    Desktop editorGitNode or PythonModel access

    Ready-to-use prompt ↓

    Open https://github.com/MrChartist/memvault. Clone the repository, inspect the README first, then set up the project locally using the exact setup steps from the README. Before editing anything, explain the stack, folder structure, data flow, required environment variables, and the safest first customization for this project: Run node init.mjs to configure, npm start to launch the web UI, then npm run sync:all to populate data — then connect your AI client and test vault_smart_search.

    How it works: Paste the prompt into Antigravity after choosing the repo. Ask it to read the README first, run setup commands, keep changes small, and explain every file it touches before editing.

    OpenAI Codex

    Best when you want command-by-command setup, debugging, code edits, and verification in a local workspace.

    GitTerminalNode or PythonLocal workspace

    Ready-to-use prompt ↓

    I want to work on https://github.com/MrChartist/memvault. First read the README and summarize what the project does. Then give me the exact commands for my operating system to clone, install, configure environment variables, and run it locally. After it runs, inspect the codebase and propose one small change that matches this goal: Run node init.mjs to configure, npm start to launch the web UI, then npm run sync:all to populate data — then connect your AI client and test vault_smart_search.

    How it works: Paste the prompt into Codex with the repository URL. Ask it to run the project, identify the stack from files instead of guessing, make scoped edits, and verify with build or browser checks.

    Claude

    Best when you want architecture explanation, codebase understanding, and a clean implementation plan before edits.

    Repo accessREADME contextTerminal if using Claude Code

    Ready-to-use prompt ↓

    Analyze https://github.com/MrChartist/memvault from the README and source structure. Create a clear architecture map, explain the main modules, identify setup risks, and recommend a safe implementation plan for extending it. Focus especially on: Ask the AI to map the MCP server architecture: mcp-server.mjs (20 tools, 5 resources, 3 prompts) → vault.mjs (SQLite CRUD) → context-engine.mjs (auto-tag, score, dedup) → ai-engine.mjs (Gemini integration). Ask it to trace a specific sync engine: e.g., sync-git.mjs scans gitDirs from config → runs git log → inserts commit entries into the vault with auto-tagging. Ask it to explain the Smart Context Engine in context-engine.mjs: how entries are auto-tagged, relevance-scored, deduplicated, and surfaced via vault_smart_context. Ask it to trace the AI importer pipeline: e.g., import-chatgpt.mjs reads conversations.json → parses messages → inserts each exchange as a vault entry. Start with one focused change: add a new sync engine (e.g., Notion or Obsidian), create a custom MCP tool for your workflow, or tune the auto-tagging rules.

    How it works: Paste the prompt with the GitHub link. Ask Claude to read the README, map modules, list risks, and produce a practical plan before using Claude Code or another agent to implement.

    AI Focus

    What to Ask the AI.

    After setup, use these focused questions to get the most value from your AI assistant.

    AI Briefing Mode

    Ask for maps, not magic.

    Use these prompts after the repo runs locally. The goal is to make the AI trace the system before it changes code.

    FlowLogicRiskChange
    01

    Explain the architecture

    Ask the AI to map the MCP server architecture: mcp-server.mjs (20 tools, 5 resources, 3 prompts) → vault.mjs (SQLite CRUD) → context-engine.mjs (auto-tag, score, dedup) → ai-engine.mjs (Gemini integration).

    02

    Prompt 2

    Ask it to trace a specific sync engine: e.g., sync-git.mjs scans gitDirs from config → runs git log → inserts commit entries into the vault with auto-tagging.

    03

    Prompt 3

    Ask it to explain the Smart Context Engine in context-engine.mjs: how entries are auto-tagged, relevance-scored, deduplicated, and surfaced via vault_smart_context.

    04

    Prompt 4

    Ask it to trace the AI importer pipeline: e.g., import-chatgpt.mjs reads conversations.json → parses messages → inserts each exchange as a vault entry.

    05

    Make one focused change

    Start with one focused change: add a new sync engine (e.g., Notion or Obsidian), create a custom MCP tool for your workflow, or tune the auto-tagging rules.

    Best Next Step

    Run node init.mjs to configure, npm start to launch the web UI, then npm run sync:all to populate data — then connect your AI client and test vault_smart_search.

    Open on GitHubBrowse All Projects

    Previous

    Newsdesk

    Next

    TV Scraper