Memex - Giving Claude a memory

@dioniipereyraa

About Memex - Giving Claude a memory

One searchable memory of your Claude.ai chats and Claude Code sessions, served over MCP. Local-first MCP server that indexes your full Claude.ai history and your Claude Code / terminal sessions, then exposes them to Claude Code (stdio) and to Claude.ai (remote connector). Hybrid

#memory #productivity #claude #claude-code

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "memex": {
      "command": "uvx",
      "args": [
        "--from",
        "memex-chats",
        "memex-mcp"
      ]
    }
  }
}

Tools

No tools detected

Fetch the live tool list by running this server in a temporary sandbox using the button above.

Overview

What is Memex?

Memex is a local-first MCP server that indexes your Claude.ai chat history and your local Claude Code/terminal sessions, making both searchable from Claude Code (via stdio) and Claude.ai (via a remote MCP connector). It solves the gap between planning on claude.ai and executing in Claude Code—Claude can now pull in past conversations without manual copy-paste.

How to use Memex?

Install Memex with a single command (curl/sh or PowerShell script), which runs memex setup to wire it into Claude Code, install the always-on capture service, and print a pairing token. Then install the Memex Chrome extension, paste the token, and click “Backfill claude.ai history.” For a manual path, clone the repo, run uv sync, then uv run memex ingest (or memex ingest from a pip install). Start the MCP server with uv run memex-mcp and add a .mcp.json config to Claude Code. Searches are also available from the CLI via memex search.

Key features of Memex

Hybrid search (vector + FTS5 BM25) across all indexed chats.
Live capture via Chrome extension for new claude.ai chats.
Local-first: data stored in SQLite + sqlite-vec, no cloud dependency.
Embeddings via fastembed (zero-config) or optional local Ollama.
Auto-summaries and chat ↔ repo association.
One-command cross-platform setup (macOS, Linux, Windows).
Remote MCP connector for claude.ai with GitHub OAuth and IP‑level tunnel.

Use cases of Memex

Ask Claude Code about a decision discussed only on claude.ai.
Search your entire chat history from Claude Code without leaving the terminal.
“Find more like this” for a current problem—find_related retrieves semantically related conversations.
Backfill and keep claude.ai history indexed while working in Claude Code.
Connect from claude.ai web or mobile to the same indexed corpus via a remote MCP connector.

FAQ from Memex

Where does my data live?

From a cloned repo, the database and exports stay in <repo>/data. From a pip/uvx install, they go to your OS data directory (~/Library/Application Support/memex on macOS, %LOCALAPPDATA%\memex on Windows, ~/.local/share/memex elsewhere). Override with MEMEX_DB_PATH / MEMEX_EXPORTS_DIR.

Do I need an API key or Ollama for embeddings?

No. The default uses a quantized 130 MB fastembed model downloaded automatically on first ingest. To use a local Ollama instead, set MEMEX_EMBED_BEDEND=ollama and pull nomic-embed-text.

How do I connect Memex to claude.ai as a remote MCP server?

Run memex serve-remote and make it publicly reachable (e.g., via Tailscale Funnel). Then create a GitHub OAuth App, restrict access to an allow‑list of GitHub usernames, and paste the server URL into claude.ai’s custom connector setup. One connector works across web, desktop, and mobile.

What tools does the MCP server expose?

Four tools: search_chats, get_chat, list_recent_chats, and find_related. The same functionality is available from the CLI with memex search, memex stats, etc.

How do I run diagnostics if something isn’t working?

Run memex doctor. It checks Python version, database health, embedder, live‑capture server, summarizer config, registered repos, and indexed corpus, reporting OK / WARN / FAIL per check.

Frequently asked questions

Where does my data live?

From a cloned repo, the database and exports stay in `<repo>/data`. From a pip/uvx install, they go to your OS data directory (`~/Library/Application Support/memex` on macOS, `%LOCALAPPDATA%\memex` on Windows, `~/.local/share/memex` elsewhere). Override with `MEMEX_DB_PATH` / `MEMEX_EXPORTS_DIR`.

Do I need an API key or Ollama for embeddings?

No. The default uses a quantized 130 MB fastembed model downloaded automatically on first ingest. To use a local Ollama instead, set `MEMEX_EMBED_BEDEND=ollama` and pull `nomic-embed-text`.

How do I connect Memex to claude.ai as a remote MCP server?

Run `memex serve-remote` and make it publicly reachable (e.g., via Tailscale Funnel). Then create a GitHub OAuth App, restrict access to an allow‑list of GitHub usernames, and paste the server URL into claude.ai’s custom connector setup. One connector works across web, desktop, and mobile.

What tools does the MCP server expose?

Four tools: `search_chats`, `get_chat`, `list_recent_chats`, and `find_related`. The same functionality is available from the CLI with `memex search`, `memex stats`, etc.

How do I run diagnostics if something isn’t working?

Run `memex doctor`. It checks Python version, database health, embedder, live‑capture server, summarizer config, registered repos, and indexed corpus, reporting OK / WARN / FAIL per check.