MCP 服务器

1AIVault

Portable AI memory vault — save memories, skills, and configs to a local database and share them across Claude Desktop, Claude Code, Cursor, Cline, and more.

README

<h1 align="center">1AIVault — MCP Server</h1>

Your portable AI memory vault — memories, skills & configs, shared across every AI tool.

1AIVault is a local-first desktop app that acts as a portable memory vault for AI tools. It stores your memories, decisions, preferences, and reusable skills in a local SQLite database and exposes them to your AI clients through this Model Context Protocol (MCP) server — so what you tell Claude Desktop is available in Cursor, Claude Code, Cline, and more. No cloud, no account, no data leaving your machine.

Website & download: https://1aivault.com

Why

Every AI tool has its own memory silo. Context you build up in one is invisible to the next. 1AIVault gives you one vault that all your AI clients read and write through a single MCP server — a shared, portable brain you own, stored on your own disk.

🧠 Cross-tool memory — save in Claude Desktop, recall in Cursor
🔒 Local-first — a single SQLite DB at ~/.1aivault/vault.db, WAL mode, never uploaded
🕸️ Knowledge graph — link entries, traverse related context, auto-extract topics across sources
🧩 Reusable skills — save AI instructions once, load them by name anywhere
⚡ Zero-config install — the app writes the MCP config for each client in one click

Install

1AIVault is a desktop app that bundles this MCP server — there is nothing to npx or clone.

Download the app for your OS from https://1aivault.com (macOS · Windows · Linux)
Open 1AIVault
Go to Connect and click Install next to your AI client — 1AIVault writes the MCP config automatically

Supported AI clients

Client	Auto-install
Claude Desktop	✅
Claude Code	✅
Cursor	✅
Cline (VS Code)	✅
Codex CLI	✅
Gemini CLI	✅
OpenCode	✅
Antigravity CLI	✅
ChatGPT Desktop	✅

Any MCP-compatible client that supports stdio servers can also be pointed at it manually (see How it works).

Tools

The server exposes 20 tools over stdio. Your AI client calls them automatically as you work.

Tool	What it does
`vault_save`	Save a memory, decision, preference, or fact to the persistent vault
`vault_search`	Full-text search the vault — supports `"quoted phrases"`, `-exclude`, `tag:`, `tier:`, `category:`
`vault_recent`	List recently saved entries, chronologically
`vault_by_tag`	Fetch entries by tag
`vault_update`	Edit or rename an existing entry
`vault_tags`	List all tags with counts
`vault_get`	Fetch one entry by ID (full content)
`vault_save_conversation`	Archive the current conversation as a summary + full text
`vault_load_skill`	Load a user-defined skill (reusable AI instructions) by name
`vault_list_skills`	List saved skills (name + description)
`vault_link`	Link two entries with a named relationship (knowledge-graph edge)
`vault_related`	Walk the knowledge graph from an entry (up to 3 hops)
`vault_brain_query`	Keyword search + graph traversal to surface richer connected context
`vault_extract_insights`	Run auto-linking / pattern extraction over entries
`vault_classify_pending`	Fetch entries not yet classified into topics
`vault_save_classifications`	Assign entries to cross-source topics (and update digests)
`vault_list_topics`	List cross-source topics with entry counts and digest previews
`vault_topic_digest`	Get a topic's cross-source summary, aliases, and entries
`vault_search_topics`	Full-text search across topics
`vault_link_topics`	Declare typed relationships between topics

How it works

Three processes share one SQLite database (~/.1aivault/vault.db, WAL mode) — no HTTP bridge, WAL handles cross-process concurrency:

Main (Electron) ─┐                                ┌─ MCP server subprocesses
                 ├── ~/.1aivault/vault.db (WAL) ──┤   (spawned by Claude Desktop,
Renderer (UI) ───┘                                └─   Cursor, Claude Code, Cline, …)

Each AI client spawns its own copy of the MCP server as a stdio subprocess. On a packaged install, the server runs through the app's own runtime (so it shares the exact native-module ABI shipped with the app), and 1AIVault writes a config like this for you — you don't edit it by hand:

<details> <summary>Example generated config (macOS)</summary>

{
  "mcpServers": {
    "1aivault": {
      "command": "/Applications/1AIVault.app/Contents/MacOS/1AIVault",
      "args": [
        ".../app.asar.unpacked/dist/main/main/mcp/server.js",
        "--source", "claude_desktop",
        "--db", "/Users/you/.1aivault/vault.db"
      ],
      "env": {
        "ELECTRON_RUN_AS_NODE": "1",
        "NODE_PATH": ".../app.asar.unpacked/node_modules"
      }
    }
  }
}

Paths are resolved per-machine by the app. The --source tag records which client wrote an entry; --db points at your vault. </details>

Privacy

100% local. Your vault is a file on your disk. Nothing is sent to any server by the MCP layer.
You own the data. Curate, export, and move it between machines from the desktop app.
No account required to use the vault or the MCP server.