MCP.so
ログイン
サーバー
A

Agentmemory

@Obidel

MCP-native memory layer for Claude Code, Cursor, Cline, Continue, and 16 other AI tools. Hybrid search (BM25 + pgvector + graph), self-hosted on Supabase + Vercel, 100% MIT, no paywall.

概要

What is Agentmemory?

Agentmemory is a visual memory manager for AI agents, providing one source of truth for tools like Claude Code, Cursor, Cline, and Continue. It is an MCP-native, open-source (MIT) server that stores and retrieves agent memories with a graph view, semantic search, and project organization.

How to use Agentmemory?

Clone the repo and run npm install then npm run dev for the web UI (local mode), or npm run mcp to start the MCP server. Configure the MCP server in your AI tool’s settings (e.g., claude_desktop_config.json) with the command npx tsx /absolute/path/to/agentmemory/mcp/index.ts. Optionally sign in for cloud sync via Supabase.

Key features of Agentmemory

  • Visual memory graph with semantic relations
  • 9 MCP tools (add, search, list, find similar, delete, etc.)
  • 3 MCP resources (rules, graph, projects)
  • Supports 20+ AI tools (Claude, Cursor, Cline, etc.)
  • Self-hostable, no telemetry, no lock-in
  • Import/export to CLAUDE.md, .cursorrules, MemGPT JSON

Use cases of Agentmemory

  • Maintain persistent, searchable memory across AI agent sessions
  • Share context between different AI tools (e.g., Claude Desktop and Cursor)
  • Visualize and explore memory relationships with a force-directed graph
  • Organize memories by project for multi-project workflows
  • Sync memories across devices via optional cloud backend

FAQ from Agentmemory

What makes Agentmemory different from simple file-based memory?

It offers a visual graph, semantic search via vector embeddings, project separation, and MCP-native integration with many AI tools, all while being fully self-hostable and free.

What are the runtime requirements for Agentmemory?

Requires Node.js and npm. The MCP server runs via tsx with no build step. For the web UI, a modern browser suffices. Cloud sync requires a Supabase account (optional).

Where are memories stored by default?

In local-only mode, memories are stored in the browser’s localStorage for the web UI, or in ~/.agentmemory/ for the stdio MCP server. Cloud sync stores data in Supabase with row-level security.

Does Agentmemory have any limits or paid tiers?

No. It is 100% free forever with unlimited memories and projects, and no premium tiers or paywalls.

How do I authenticate for cloud sync?

Sign in to the hosted web UI, then retrieve a Supabase JWT from DevTools (localStorage.getItem('sb-<project>-auth-token')). The JWT expires after ~1 hour; refresh by re-running the command.

タグ

「開発者ツール」の他のコンテンツ