Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

cliwant/agent-tools

Open-source primitives for AI-native applications. Each package is shipped as both an MCP server and a Claude Code skill, so you can wire it into Claude / Codex / Gemini / Cursor / Cline with one line of config.

Status: private preview — the repos are private until the maintainers flips them public. If you got a link, please don't share it yet.

Packages

Package Form Status Purpose
@cliwant/mcp-memory-tiers MCP + Skill ✅ v0.1 Three-tier persistent memory (digest / working_set / long_term vector).
@cliwant/mcp-llm-gateway MCP + Skill ✅ v0.1 Vendor-agnostic LLM gateway (Anthropic / OpenRouter / OpenAI / CLI) with credit-balance-aware fallback chain.
@cliwant/mcp-approval-queue MCP + Skill ✅ v0.1 Human-in-the-loop approval queue for tool-using agents (queue / approve / modify / wait).

Each package has independent value — you can install one without the others.

Quick start

# 1. Install the package globally (or run via npx)
npm install -g @cliwant/mcp-memory-tiers

# 2. Add it to your MCP host's config
#    (see ./packages/mcp-memory-tiers/README.md for per-host snippets)

# 3. Restart your host. The eight tools become callable from Claude / Codex / Gemini.

Cross-platform

Platform Verified Notes
macOS (Intel + Apple Silicon) Node 20+ via Homebrew or fnm.
Windows 11 (incl. ARM64) Node 20+ via fnm or the official installer. better-sqlite3 ships prebuilt for win32-x64 and win32-arm64.
Linux (Ubuntu 22.04+) Node 20+.

See docs/install.md for the per-platform install walkthrough.

Layout

agent-tools/
├── packages/
│   ├── mcp-memory-tiers/   ← shipped MCP + Skill, v0.1
│   ├── mcp-llm-gateway/    ← stub
│   └── mcp-approval-queue/ ← stub
├── skills/
│   ├── cliwant-memory-tiers/    SKILL.md
│   ├── cliwant-llm-gateway/     SKILL.md (stub)
│   └── cliwant-approval-queue/  SKILL.md (stub)
├── docs/
│   ├── install.md
│   └── architecture.md
└── README.md

Each packages/<name>/ ships independently to npm. The skills/<name>/ directories are copy-into-~/.claude/skills/ overlays for hosts that support skill-format declaratives on top of the MCP servers.

Design principles

  1. One job per package. Memory is one package. LLM routing is another. Approval queue is a third. Composing them is the host's call, not ours.
  2. Standards over inventions. We speak Model Context Protocol verbatim. We don't invent new wire formats.
  3. Zero-config defaults. SQLite, no embedding key, ENOENT-safe paths — the package runs on a fresh machine without an external dependency.
  4. Privacy-by-default. Storage is local. Nothing phones home. Topic isolation is enforced at the storage layer.
  5. i18n in shipping. README + skill manifest in EN / 한국어 / 日本語 from day one.

License

MIT for every package. See each package's LICENSE file.

Maintainer

seungdo-keum — currently working on the Practiq AI-native workspace for boutique professional services firms. This monorepo is the OSS substrate of the same primitives.

About

Open-source MCP servers + Claude Code skills for AI-native applications. Memory tiers, multi-provider LLM gateway, approval queue. Works with Claude / Codex / Gemini / Cursor / Cline.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages