LLMemory MCP Server

A remote MCP server that lets AI coding agents search and pull context from your imported AI chat history (ChatGPT, Claude, Codex, Claude Code) without copy-pasting. Reads your vault live from your own Google Drive. Zero server-side storage, fully private.

Server URL: https://mcp.llmemory.xyz/mcp Transport: Streamable HTTP Auth: OAuth 2.1 with Dynamic Client Registration (no API keys — browser flow on first use)

Install in Cline

This is a remote MCP server — no cloning, no npm install, no local process. Cline just needs the URL.

Option A — Cline Dashboard (recommended):

1. Log in to the Cline Dashboard → Organization → MCP Servers
2. Add a new Remote MCP server
3. Name: LLMemory · URL: https://mcp.llmemory.xyz/mcp
4. Save. In VS Code/JetBrains, open the Cline panel → MCP Servers → connect llmemory.
5. On first tool call, a browser window opens — authorize with Google and grant Drive access.

Option B — Cline MCP config JSON: Add this to your Cline MCP settings (cline_mcp_settings.json):

{
  "mcpServers": {
    "llmemory": {
      "type": "remote",
      "url": "https://mcp.llmemory.xyz/mcp"
    }
  }
}

Restart the Cline extension. On first tool call, the OAuth browser flow opens automatically.

Install in other MCP clients

Same URL works in any MCP client that supports Streamable HTTP + OAuth 2.1:

• Cursor — LLMemory Cursor plugin
• VS Code / GitHub Copilot — add to .vscode/mcp.json: { "servers": { "llmemory": { "type": "http", "url": "https://mcp.llmemory.xyz/mcp" } } }
• Claude Desktop / Claude Code — claude mcp add --transport http llmemory https://mcp.llmemory.xyz/mcp
• Open WebUI — Admin Settings → External Tools → add Streamable HTTP MCP server with URL https://mcp.llmemory.xyz/mcp

Tools

Tool	Purpose
list_sources	Show which AI providers have data in your vault and counts
list_recent	Recently updated conversations (id, title, source, message count). Filter by source
search_chats	Full-text search across all imported conversations. Returns snippets + conversation IDs
get_conversation	Fetch one full transcript by ID (Markdown or JSON)
get_context_pack	Agent-ready briefing (goal, recent state, files, open tasks) for one conversation or a whole folder

What it reads

LLMemory reads the LLMemory/ folder in your Google Drive — a manifest.json plus one JSON file per conversation, written by the LLMemory browser extension or CLI. The server is read-only and zero-storage: it brokers Google OAuth, streams the result, then forgets everything. Your chats never leave your Drive on a server.

Privacy

• Zero server-side storage. No chat content, no per-user database. Cloudflare Workers free tier.
• Your Drive is the storage. Chats live in your own Google Drive LLMemory/ folder.
• Read-only. The connector never writes to your vault.
• OAuth-scoped. The server only ever sees your LLMemory/ Drive folder (drive.file scope).

Full privacy posture: STORE_PRIVACY.md

Examples

• "What did I figure out about the OAuth worker last week?" → searches your vault
• "Pull my most recent Codex session" → list_recent filtered to codex, then get_conversation
• "Give me a context pack for the session where we built the Claude connector" → get_context_pack
• "Show me which chats have data imported" → list_sources

Links

• LLMemory
• MCP connector design doc
• Main repo

License

MIT