Idea: optional Vestige memory backend as a Memory MCP component (keeps the same tool contract)

[samvallad33]

Disclosure first: I maintain Vestige, so this is an interested-party idea. Per CONTRIBUTING I am raising it as a Discussion before any PR.

UltraRAG 2.0 makes each block its own MCP server (Corpus, Retriever, Generation, Router, Note), which is a really clean design. Memory today is within-session, so iterative pipelines like Search-o1, WebNote, and Iter-RetGen do not carry knowledge across runs. Because everything is already MCP, a memory server can slot in as an optional Memory component with no glue.

Vestige - Local-first cognitive memory MCP server for AI coding agents. FSRS-6 retention, prediction-error gating, active forgetting, spreading activation, 3D dashboard. Single Rust binary. npm install -g vestige-mcp-server.

It would sit alongside your existing blocks as a Memory server and give pipelines durable cross-run recall that survives restarts. I am happy to contribute a small example pipeline YAML showing memory-augmented multi-turn RAG if the direction is interesting. Storage is local by default, which fits an academic and open-source setup.

---

（中文）先说明：我是 Vestige 的作者，属于利益相关方。按 CONTRIBUTING 的要求，先在 Discussion 里提想法，再考虑 PR。

UltraRAG 2.0 把每个模块都做成独立的 MCP server（Corpus、Retriever、Generation、Router、Note），这个设计很干净。目前记忆是会话内的，Search-o1、WebNote、Iter-RetGen 这类迭代式 pipeline 没法跨运行保留知识。既然整体已经是 MCP，一个记忆服务可以作为可选的 Memory 组件接入，几乎不需要胶水代码。

Vestige 是本地优先的认知记忆 MCP 服务：FSRS-6 留存、预测误差门控、主动遗忘、spreading activation、3D dashboard，单个 Rust 二进制。它可以和现有模块并列，作为 Memory server，给 pipeline 提供能在重启后保留的长期记忆。如果方向可行，我很乐意贡献一个 memory-augmented 多轮 RAG 的示例 pipeline YAML。默认数据全部本地存储，比较契合学术和开源场景。

Sam Valladares, github samvallad33