tiny-memory-mcp

Tiny Memory MCP

A small, dependency-free MCP server written in Python. It stores durable memories in a local SQLite database and exposes them as MCP tools over stdio or Streamable HTTP.

Tools

• search: search memories in ChatGPT connector format
• fetch: fetch one memory by key in ChatGPT connector format
• remember: store or update a memory by key
• recall: search memories by key, value, or tag
• list_memories: list recent memories
• forget: delete a memory by key

Run Locally With Stdio

python3 server.py

The server speaks MCP over stdio, so it is meant to be launched by an MCP host.

Configure In An MCP Host

Use the absolute path to this folder on your machine:

{
  "mcpServers": {
    "tiny-memory": {
      "command": "python3",
      "args": [
        "/Users/dawn/Code/ai-experiments/tiny-memory-mcp/server.py"
      ],
      "env": {
        "TINY_MEMORY_DB": "/Users/dawn/Code/ai-experiments/tiny-memory-mcp/memory.sqlite3"
      }
    }
  }
}

Manual Smoke Test

python3 tests/smoke.py

This starts the server as a subprocess, initializes an MCP session, stores a memory, searches for it, lists memories, deletes it, and verifies it is gone.

Run Locally With Streamable HTTP

HOST=127.0.0.1 PORT=8000 python3 http_server.py

The MCP endpoint is:

http://127.0.0.1:8000/mcp

HTTP smoke test:

python3 tests/http_smoke.py

Docker

Build and run:

docker build -t tiny-memory-mcp .
docker run --rm -p 8000:8000 -v tiny-memory-data:/data tiny-memory-mcp

Or with Compose:

docker compose up --build

Expose With Ngrok

Once the Docker container is running on port 8000:

ngrok http 8000

Use the HTTPS forwarding URL with /mcp appended:

https://YOUR-NGROK-DOMAIN.ngrok-free.app/mcp

In ChatGPT custom MCP settings, choose Streamable HTTP and use that URL.