Getting Started

Getting Started & Setup

Bring the MCP Local Memory Service to life from two perspectives: as a developer you install and register the server, and as an AI agent you discover and start using your new memory.

🧑‍💻 As a Developer: Setting Up the Infrastructure

1. Installation

Install the service globally on your machine:

npm install -g @vheins/local-memory-mcp

2. Configuration (User Level)

Register the server in your preferred AI tool. Below are the standard configurations:

Client	Configuration file / location
Claude Desktop	macOS ~/Library/Application Support/Claude/claude_desktop_config.json · Windows %APPDATA%\Claude\claude_desktop_config.json
Cursor	~/.cursor/mcp.json
Windsurf	~/.codeium/windsurf/mcp_config.json
Gemini CLI / Antigravity	~/.gemini/settings.json (or .gemini/settings.json in project root)
Trae	Settings > MCP > Add Manually
Kiro AI	MCP icon in the sidebar → add to mcp-servers.json
Zed	~/.config/zed/settings.json (context_servers, see below)
OpenCode	Extension settings, using the standard npx command and arguments

Example for Gemini CLI / Antigravity:

{
	"mcpServers": {
		"local-memory": {
			"command": "npx",
			"args": ["-y", "@vheins/local-memory-mcp"]
		}
	}
}

Example for Zed:

"context_servers": {
  "local-memory": {
    "command": {
      "path": "npx",
      "args": ["-y", "@vheins/local-memory-mcp"]
    }
  }
}

Dedicated setup guides: Claude Code · Codex · Kiro.

🤖 As an AI Agent: Discovering Your New Memory

Once the developer has registered the server, the agent takes over the integration process:

1. Handshake & discovery — when a session starts, the agent performs a handshake with the MCP server and discovers the 17 unified tools (memory-write, memory-read, task-write, task-read, standard-read, codebase-index, codebase-read, claim-manage, handoff-write, and more), resource templates such as repository://{name}/memories, and the memory-agent-core prompt that defines its behavioral contract.
2. Context activation — the agent resolves the current Git Repository name to isolate memories, and can pull relevant memories from other projects via tags (Tech-Stack Affinity) — searching with current_tags (e.g. ["react", "filament"]) surfaces knowledge tagged with those technologies from any repo.
3. Validation ritual — ask: "What do you remember about this project?" or "Check your local memory for any coding standards related to this tech stack." If the agent responds with a list of facts or a "No memories found yet" message (instead of an error), you are successfully integrated.

⚠️ Disclaimer

THE SOFTWARE IS PROVIDED "AS IS", without warranty of any kind. Both developer and agent must collaborate to ensure the quality and accuracy of stored data.

Related Pages

• Tool Reference & Usage Guide — the 17 tools with usage examples
• MCP Concepts — protocol-level interface
• Web Dashboard Guide — visual management of your data
• Hybrid Search — how search ranks results
• Troubleshooting Guide — common issues and fixes
• Memulai & Pengaturan — Bahasa Indonesia