Memory Store

Memory Store

A Rust MCP stdio server for durable project-local AI notes, enabling agents to store, search, and manage verified facts, decisions, and conventions.

Category
访问服务器

README

Memory Store

Memory Store is a small Rust MCP stdio server for durable project-local AI notes. It gives agents a place to remember verified project facts, decisions, setup commands, debugging results, and conventions without hiding them in a global database or editing random scratch files.

The canonical store is plaintext and git-friendly:

.memory-store/
  metadata.toml
  VERSION
  .gitignore
  memories/
    <cuid2>.md
  .cache/
    fff/

Each memory is a Markdown file with TOML frontmatter and a generated Cuid2 key. Search uses FFF over .memory-store/memories. The .memory-store/.cache/fff directory is disposable search/index state and is ignored by git.

Use Case

Use Memory Store for knowledge that should survive between AI sessions but still belong to the project:

  • repo setup facts that were actually verified
  • architecture or product decisions with short rationale
  • debugging outcomes, exact symptoms, and confirmed fixes
  • conventions future agents should follow
  • small reminders that are too local for global memory and too operational for committed docs

Do not store secrets, credentials, personal data, long logs, full docs, copied source files, or guesses that have not been verified.

Tools

Memory Store exposes these MCP tools:

  • init {}: create .memory-store and metadata for the current project.
  • new { content, tags? }: create a new memory with a generated Cuid2 key.
  • get { key }: read one memory by opaque key.
  • update { key, expected_revision, operations }: apply exact body edits after checking the current revision.
  • delete { key, expected_revision }: remove a memory after checking the current revision.
  • search { query, limit?, offset? }: search memories with FFF-backed content and fuzzy search.

new, get, update, delete, and search intentionally fail until the project has been initialized:

MemoryStore is not initialized in this project. Initialize it first via the `init` command instead.

Install For Humans

Memory Store currently builds the Rust binary from source when installed from npm, so you need a Rust toolchain with cargo on PATH.

Install globally:

npm install -g @barekey/memorystore

Or run it through npx from an MCP client:

[mcp_servers.memorystore]
command = "npx"
args = ["-y", "@barekey/memorystore"]
startup_timeout_sec = 30

From a source checkout:

cargo build --release --bin memorystore-mcp

Then point your MCP client at the built binary:

[mcp_servers.memorystore]
command = "/path/to/memorystore/target/release/memorystore-mcp"
args = []
startup_timeout_sec = 30

On Windows, use the .exe path.

To initialize a project outside an MCP client:

memorystore-mcp --root /path/to/project init

Install The Skill

This repository also contains a Codex skill that teaches agents how to use the MCP server safely.

Install it with the open skills CLI:

npx -y skills add usebarekey/memorystore --skill memorystore --full-depth

After installing, invoke it in Codex with:

Use $memorystore to initialize, search, update, and clean up durable project memories for this task.

The skill tells agents to use MCP tools only. They should not manually read, write, edit, rename, migrate, or delete files inside .memory-store.

Memory Shape

Keep notes small and focused. A good memory is usually one paragraph: one fact, decision, convention, or debugging result. If a note needs multiple sections or long excerpts, split it into smaller memories or move it into committed project docs.

Keys are opaque Cuid2 identifiers. Do not use keys as taxonomy. Search by topic, tags, component names, error text, and the note body, then use returned keys only for exact get, update, or delete calls.

Development

Run the Rust checks:

cargo fmt --check
cargo clippy --all-targets -- -D warnings
cargo test

Check the npm package contents:

npm pack --dry-run

Validate the bundled skill:

python C:/Users/Sander/.codex/skills/.system/skill-creator/scripts/quick_validate.py skills/memorystore

Release

Create the GitHub repository under the usebarekey organization and push:

gh repo create usebarekey/memorystore --public --source . --remote origin --push

Publish the npm package:

npm publish --access public

The npm package is scoped as @barekey/memorystore and exposes the memorystore-mcp and memorystore binaries.

License

BSD 3-Clause. See LICENSE.

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选