自托管 MCP 记忆服务器 recall：跨模型共享 agent 记忆
一个自托管 MCP 服务器，让不同 agent 把记忆写入同一处并互相读取，作者已接入 Meta Muse 和 Grok；仅限本地网络或 Tailscale 使用。
I think memory with agents should be model-agnostic. a little over a year ago, I realised that a major lock-in of tools like ChatGPT, Claude and Gemini is that they slowly get to know you and that the utility of them having personalised replies keeps you in their apps. I have tried to build out small self-hosted MCP servers before, which act as a place for agents to dump their memories and go back there for recall, but it's never worked well. Opus 5.5 just pretty much one-shotted it and it works well. I've had my Meta @Muse dump its memories, as well as my Grok @Bot and they can access one anothers' memories! do NOT host this anywhere public. if you're trying this out, only do it on a local network, or something like a Tailscale Tailnet. https://github.com/j4ckxyz/recall