By MEOK AI Labs — Sovereign AI tools for everyone.
Memory Search MCP Server
Persistent memory system for AI agents and assistants. Record episodic memories with care-weighting and emotional valence, search with full-text relevance ranking, maintain a knowledge base of facts and reference material, follow temporal chains, and consolidate old memories automatically.
Zero external dependencies beyond the mcp package -- uses SQLite with FTS5 for fast full-text search. Data persists in ~/.mcp-memory/memories.db.
Tools
| Tool | Description |
|---|---|
record_memory |
Store a memory episode with care weight, importance, emotion, and tags |
search_memory |
Full-text semantic search with care weight and tag filtering |
add_knowledge |
Add persistent facts/reference material to the knowledge base |
search_knowledge |
Search the knowledge base by topic or content |
list_memories |
Browse recent memories sorted by time, importance, or access count |
get_memory_stats |
Memory store statistics: counts, averages, storage size |
get_temporal_chain |
Follow the timeline forward/backward from any memory |
consolidate_memories |
Archive old low-access memories to save space |
Installation
pip install mcp
Usage
Run the server
python server.py
Claude Desktop config
{
"mcpServers": {
"memory-search": {
"command": "python",
"args": ["/path/to/memory-search-mcp/server.py"]
}
}
}
Example calls
Record a memory:
Tool: record_memory
Input: {"content": "User prefers dark mode and compact layouts", "source_agent": "preferences", "memory_type": "insight", "care_weight": 0.8, "tags": ["preferences", "ui"]}
Output: {"success": true, "episode_id": "a3f2b1c8d9e0", "timestamp": "2026-04-13T10:30:00"}
Search memories:
Tool: search_memory
Input: {"query": "user interface preferences", "limit": 5, "care_weight_min": 0.5}
Output: {"results": [...], "count": 3, "query": "user interface preferences"}
Add knowledge:
Tool: add_knowledge
Input: {"topic": "Python asyncio", "content": "Use asyncio.gather() for concurrent coroutines...", "confidence": 0.9}
Output: {"success": true, "knowledge_id": "k1a2b3c4d5e6", "topic": "Python asyncio"}
Follow temporal chain:
Tool: get_temporal_chain
Input: {"episode_id": "a3f2b1c8d9e0", "direction": "backward", "max_steps": 10}
Output: {"chain": [...], "direction": "backward", "steps": 7}
Data Storage
All data is stored in ~/.mcp-memory/memories.db (SQLite). To reset, simply delete this file. To back up, copy it.
Pricing
| Tier | Limit | Price |
|---|---|---|
| Free | 100 calls/day | $0 |
| Pro | Unlimited + vector embeddings + cloud sync | $9/mo |
| Enterprise | Custom + team sharing + encryption at rest | Contact us |
License
MIT
