跳到正文
TakoAPI
返回技能列表

clude-memory-mcp

由 sebbsssss 提供

Git & GitHub
284下载量由 clawskills.sh 提供
0浏览量
0 个赞登录后即可为这个技能点赞
—暂无评分
你的评分

简介

MCP server for Clude's 4-tier cognitive memory system — store, recall, search, and dream.

描述

该技能的功能

MCP server that gives agents a 4-tier memory system: episodic, semantic, procedural, and self-model. Each tier decays at its own daily rate, and memories are retrieved using hybrid vector and keyword search. Memories are committed on-chain to Solana as SHA-256 hashed memo transactions.

Unlike flat key-value memory stores, it applies type-specific decay and hybrid retrieval combining vector similarity, keyword matching, and tag scoring, so the most relevant memories surface even across long time spans.

何时使用

  • Persisting user preferences across separate agent conversations
  • Building an agent that recalls past interactions with specific users
  • Tracking learned behaviors and procedural knowledge over time
  • Searching prior conversation history by topic or tag
  • Giving an agent a stable self-model that evolves with experience

示例工作流

输入

User asks: 'What did we discuss last week about the deployment pipeline?'

智能体
  1. 1

    Call recall_memories with query 'deployment pipeline' and memory_types ['episodic', 'semantic']

  2. 2

    Score results by relevance, recency, and importance using the hybrid retrieval system

  3. 3

    Filter by min_importance threshold to drop low-signal entries

  4. 4

    Return top-ranked memory entries to the agent context

输出

Agent surfaces relevant memory entries from the prior conversation and uses them to answer the user's question about the deployment pipeline

前置要求

  • SUPABASE_URL environment variable
  • SUPABASE_SERVICE_KEY environment variable
  • Supabase account with a project initialized using supabase-schema.sql

安装

clawhub install sebbsssss/clude-memory-mcp

运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。

链接