HF
Memory MCP

Your memories. Your graph. Your context.

Persistent structured memory for LLMs on Cloudflare Workers. Knowledge graphs, semantic search, and temporally-decayed recall — so your AI never forgets what matters.

1024-d
vector embeddings
14
MCP tools
7-day
decay half-life
BFS
graph traversal
Memory MCP ChatGPT Memory LangChain Memory Neo4j
Knowledge graph
Semantic vector search
Temporal decay ranking
Graph + vector unified
No servers to run
MCP protocol native
Multi-tenant namespaces
Context-enriched search
Conversation storage
Typed memory fragments
Weighted relations
REST + MCP dual interface
Cloudflare Zero Trust
Self-hosted
Free at any scale
1
Deploy to your Cloudflare account
One deploy creates the Worker, D1 database, Vectorize index, and Durable Objects. Everything runs on your account.
$ git clone https://github.com/FlarelyLegal/memory-mcp.git
$ cd memory-mcp
$ npx wrangler deploy
Deployed to https://memory.homeflare.dev
2
Configure Cloudflare Access
Set up an Access application with your identity provider. OAuth handles auth between your MCP client and the server automatically.
# Set Access environment variables
$ npx wrangler secret put ACCESS_CLIENT_ID
$ npx wrangler secret put ACCESS_CLIENT_SECRET
Secrets configured
3
Connect your MCP client
Point Claude Desktop, Cursor, or any MCP-compatible client to your server. Authenticate once, then your AI has persistent memory.
# claude_desktop_config.json
{
"mcpServers": {
"memory": {
"url": "https://memory.homeflare.dev/sse"
}
}
}
4
Build knowledge that persists
Create entities, link them with relations, store memories with importance scores. Semantic search and temporal decay handle the rest.
# Your AI can now remember
Created entity "Tim" (person)
Created relation Tim → uses → Cloudflare Workers
Stored memory (fact, importance: 0.9)
"Tim prefers short meetings"
Claude Desktop
Cursor
REST API
↓ MCP / HTTPS ↓
Cloudflare Edge — DDoS · TLS · Access (IdP + Zero Trust)
↓ OAuth ↓
Worker — MCP Agent → Embed → Graph ops → Decay rank → Search
↓ structured + vector ↓
D1 (graph)
Vectorize (embeddings)
Workers AI