Agent Memory Hub

Persistent, intelligent, searchable long-term memory for AI agents.

Store facts, preferences, notes, and project context. Retrieve them with full-text BM25 search, importance scoring, and recency weighting. No API keys. No external servers. Works out of the box.

---

Features

• 7 powerful tools — store, search, retrieve context, update, list, forget, summarize
• BM25 full-text search — proper ranked search with IDF, not just string matching
• Auto-tagging — automatically infers categories (preference, project, technical, task, credential, etc.)
• Auto importance scoring — detects urgency signals in content
• Recency + importance weighting — more relevant memories surface first
• Atomic writes — corruption-safe file persistence
• Zero dependencies — only the MCP SDK; no native binaries, no Python, no Docker
• Configurable storage — override path with AGENT_MEMORY_DIR env var

---

Installation

1. Clone and build

git clone https://github.com/yourname/agent-memory-hub
cd agent-memory-hub
npm install
npm run build

2. Add to Claude Desktop

Edit %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "agent-memory-hub": {
      "command": "node",
      "args": ["C:\\Users\\HP\\agent-memory-hub\\build\\index.js"]
    }
  }
}

3. Add to Claude Code (MCP CLI)

claude mcp add agent-memory-hub -- node "C:\Users\HP\agent-memory-hub\build\index.js"

Custom storage directory

{
  "mcpServers": {
    "agent-memory-hub": {
      "command": "node",
      "args": ["C:\\Users\\HP\\agent-memory-hub\\build\\index.js"],
      "env": {
        "AGENT_MEMORY_DIR": "C:\\Users\\HP\\my-agent-memories"
      }
    }
  }
}

Default storage: ~/.agent-memory/memories.json

---

Tools

store_memory

Store any piece of information worth remembering.

key:        "user_preferred_language"
content:    "User always prefers TypeScript over JavaScript"
tags:       ["preference", "technical"]   ← auto-detected if omitted
importance: 7                             ← auto-scored if omitted
overwrite:  true                          ← upsert: update if key exists, create if not

By default, storing a key that already exists returns an error. Set overwrite: true to silently update the existing memory instead — useful when you want "set this value" semantics without checking first.

search_memory

BM25 full-text search across all memories.

query: "typescript preferences"
limit: 5          ← optional, default 5
tags:  ["technical"]  ← optional filter

get_relevant_context

Auto-retrieve the best memories for a given query. Use this at session start.

user_query: "Help me set up the project authentication"
→ Returns: identity memories, project memories, technical preferences

update_memory

Modify existing memory content, tags, or importance.

key:         "user_preferred_language"
new_content: "User prefers TypeScript, but accepts Python for scripts"
importance:  8

list_memories

Browse memories with sorting and filtering.

tags: ["project"]
sort: "importance"   ← "recent" | "importance" | "access"
limit: 10

forget_memory

Permanently delete a memory.

key: "old_api_key"

memory_summary

Get a full overview — counts, top tags, most important and most accessed memories.

---

Storage Format

Memories are stored as plain JSON at ~/.agent-memory/memories.json. Human-readable, easy to backup or inspect.

{
  "version": "1.0.0",
  "created": "2025-01-01T00:00:00.000Z",
  "lastUpdated": "2025-06-01T12:00:00.000Z",
  "memories": [
    {
      "id": "uuid",
      "key": "user_preferred_language",
      "content": "User prefers TypeScript over JavaScript",
      "tags": ["preference", "technical"],
      "importance": 7,
      "createdAt": "...",
      "updatedAt": "...",
      "accessCount": 12,
      "lastAccessed": "..."
    }
  ]
}

---

Auto-Tagging Categories

The system auto-detects these categories from content:

Tag	Trigger signals
preference	prefer, like, love, hate, favorite, avoid
project	project, working on, building, repository
identity	I am, my name, I work, my role
technical	code, api, database, framework, docker
task	todo, must, deadline, remind
credential	password, secret, token, api key
note	note, remember that, fyi, heads up
person	name is, email, phone, contact
config	config, setting, env var, port, url

---

Development

npm run dev    # watch mode
npm run build  # production build

---

License

MIT