Presume I know nothing about the underlying technology and please explain how to start using this project

[FrankDMartinez]

I see an "Agent Quick Start" and a "Get Started in 60 Seconds". The first starts up just fine while the second gives an "Error while finding module specification for 'mcp_memory_service.scripts.installation.install' (ModuleNotFoundError: No module named 'mcp_memory_service.scripts')" error for Python 3.11 after performing "pip3.11 install mcp-memory-service". If I am a general Claude use, which start-up method do I want to work? If I am a Claude Code user, which start-up method do I want to work?

Thank you in advance.

[doobidoo]

Hi @FrankDMartinez, great question! Let me break this down.

The Error You Hit

The "Get Started in 60 Seconds" section tells you to run:

python -m mcp_memory_service.scripts.installation.install --quick

This fails because the scripts/ directory is not included in the PyPI package — it only exists in the git repo. So after pip install mcp-memory-service, that module path does not exist. This is a documentation bug and we will fix it.

Which Method Should You Use?

If you use Claude Desktop (the desktop app)

After installing with pip install mcp-memory-service, you need to manually configure Claude Desktop to use the memory server. Add this to your Claude Desktop config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "memory": {
      "command": "memory",
      "args": ["server"]
    }
  }
}

Then restart Claude Desktop. The memory service will start automatically when Claude Desktop launches, and you will have access to memory tools (store, search, recall) directly in your conversations.

If you use Claude Code (the CLI/terminal tool)

Claude Code uses its own MCP server configuration. Add the memory service to ~/.claude/config.json:

{
  "mcpServers": {
    "memory": {
      "command": "memory",
      "args": ["server"]
    }
  }
}

Then restart Claude Code. Memory tools will appear automatically.

If you are building agent pipelines (the "Agent Quick Start")

The Agent Quick Start is for developers building programmatic AI workflows (using LangGraph, CrewAI, custom scripts, etc.) that talk to the memory service via its REST API. You would run:

MCP_ALLOW_ANONYMOUS_ACCESS=true memory server --http

This starts an HTTP server at http://localhost:8000 that any program can call via REST endpoints. This is not needed for normal Claude Desktop or Claude Code usage.

TL;DR

You are...	Do this
Claude Desktop user	pip install mcp-memory-service then edit claude_desktop_config.json then restart
Claude Code user	pip install mcp-memory-service then edit ~/.claude/config.json then restart
Agent/developer	pip install mcp-memory-service then memory server --http then use REST API

Hope that clears things up! We will update the README to fix the broken install script reference.

[FrankDMartinez]

Thank you! BTW, how can I verify this is working? Is there a log file or something I can check?

[doobidoo]

Thanks @omni-front for chiming in! Let me give a more specific answer to @FrankDMartinez's verification question.

How to verify it's working

Option 1: CLI status check (quickest)

memory status

This connects to the local storage backend and shows version, backend type, memory count, and database size. If you see "Service is healthy" — you're good.

Option 2: Inside Claude Desktop

After configuring claude_desktop_config.json and restarting, look for the 🔨 hammer icon in the bottom-left of the chat input. Click it — you should see memory tools listed (like memory_store, memory_search, etc.). You can also just ask Claude: "Do you have access to memory tools?" — it will tell you.

Option 3: Inside Claude Code

Type /mcp in Claude Code to see connected MCP servers. The memory server should appear with its tools listed. You can also try using a memory tool directly, e.g. ask Claude to "store a test memory" — if it works, you're connected.

Option 4: HTTP health endpoint (if running the HTTP server)

curl http://127.0.0.1:8000/api/health

Returns a JSON status with backend info, memory count, and uptime.

Logs

The MCP server logs to stderr by default (which Claude Desktop captures). You won't see a separate log file unless you redirect output. For debugging, you can run the server manually in a terminal:

memory server

and watch the output — you'll see initialization messages, embedding model loading, and any errors.