Knowledge2026年5月12日·1 分钟阅读

Recall — Persistent Memory MCP for Claude Agents

Recall provides persistent cross-session memory for Claude agents via an MCP server backed by Redis/Valkey or managed cloud, with sharing and search.

Agent 就绪

这个资产可以被 Agent 直接读取和安装

TokRepo 同时提供 CLI 命令、metadata JSON、安装计划和原始内容链接,方便 Agent 判断适配度、风险和下一步动作。

Needs Confirmation · 62/100策略:需确认
目标
Claude Code
类型
Memory
安装
Manual
信任
信任等级:Established
入口
npx -y @joseairosa/recall
CLI 安装命令
npx tokrepo install 9e80e920-7c69-5688-b456-2a7c2b6f2268 --target codex

简介

Recall 是一个基于 MCP 的记忆层:把决策、需求、代码模式等跨会话保存下来,减少 Claude/Agent 每次从零开始。它支持 Redis/Valkey 自托管或托管服务,并在 README 里明确了记忆数据的安全边界。

最适合: 需要跨多次会话保留“项目决策/规范/需求”的 Agent 工作流

适配: Claude Code/Claude Desktop、Node.js 18+、Redis/Valkey(可选)、通过 MCP 客户端接入

配置时间: 2–10 分钟

关键事实(已验证)

  • README 列出托管服务定价(Free 500 memories;Pro 5K;Team 25K)。
  • README 写明自托管需要 Redis ≥6 或 Valkey,以及 Node.js ≥18。
  • GitHub:167 stars · 18 forks;最近更新 2026-04-09(GitHub API 验证)。

正文

把 Recall 当成基础设施来接入:先决定记忆数据是落在自有 Redis/Valkey 还是托管实例,然后把 MCP 命令接到你的 Agent 客户端配置里。团队使用时把记忆当作敏感数据,按 README 的建议做访问控制与加密。

上线时先从少量结构化类型(directive/decision/requirement)开始,再逐步扩展。

README 原文节选(verbatim)

Recall 🧠

Persistent cross-session memory for Claude and other AI agents. Survives context limits, session restarts, and compaction.

Store decisions, patterns, and knowledge once — retrieve them automatically across every future conversation. No more repeating yourself. No more losing critical context when the window fills up.


Cloud vs. Self-Hosted

Recall comes in two flavours. Choose the one that fits your situation:

☁️ recallmcp.com — Managed SaaS (Recommended for most users)

Zero infrastructure. Sign up, get an API key, add one config block, done.

What you get Detail
No Redis required Fully managed storage — nothing to install or maintain
Up in 2 minutes One config block in Claude Desktop or Claude Code
Team sharing One Claude learns something, all others know instantly
Enterprise security AES-256-GCM encryption at rest, per-tenant key isolation
Auto-consolidation Automatically clusters and merges similar memories
Cross-session workflows Named threads that span multiple sessions
7 embedding providers Voyage AI, Cohere, OpenAI, Deepseek, Grok, Anthropic, Ollama
Pricing Free (500 memories) · Pro $4.99/mo (5K) · Team $9.99/mo (25K)
{
  "mcpServers": {
    "recall": {
      "url": "https://recallmcp.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

Get started at recallmcp.com


🖥️ This Repository — Self-Hosted (Redis or Valkey)

Run Recall entirely on your own infrastructure. Your data never leaves your servers.

What you get Detail
Full data ownership Memories stored in your own Redis or Valkey instance
No usage limits No memory caps, no subscription required
Air-gapped deployments Works fully offline, no external API calls for storage
Bring your own embeddings Configure any of 7 providers, or use the Anthropic keyword fallback
Open source MIT licensed, fully auditable, fork-friendly
Requires Redis ≥ 6 or Valkey, Node.js ≥ 18, your own server/VPS

Best for: developers who need full control, on-premise requirements, or who already run Redis infrastructure.


⚠️ Security & Privacy Disclaimer

IMPORTANT: READ BEFORE USE

Recall stores conversation memories in either Redis or Valkey, which may contain sensitive information including:

  • Code snippets, API keys, credentials, and secrets discussed in conversations

FAQ

一定要 Redis 吗? 答:不一定。README 表格写明托管模式不需要 Redis;自托管则用 Redis/Valkey。

能不能快速开始? 答:README 提到托管模式用一段配置可在约 2 分钟上线。

最大的风险是什么? 答:记忆可能包含敏感信息;按 README 建议为 Redis/Valkey 配置鉴权、TLS 与网络隔离。

🙏

来源与感谢

Source: https://github.com/joseairosa/recall > License: MIT > GitHub stars: 167 · forks: 18

讨论

登录后参与讨论。
还没有评论,来写第一条吧。

相关资产