larder

mcp
Security Audit
Fail
Health Pass
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 29 GitHub stars
Code Fail
  • rm -rf — Recursive force deletion command in build/darwin/Taskfile.yml
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Local-first desktop app for inspecting and safely managing the Skills, MCP servers, CLIs, and instruction files your AI agents use.

README.md

Larder

Local provisions for your agents.

English · 简体中文

Larder is a local-first desktop app for developers who work with multiple AI agents. It shows what is actually installed on your machine — Skills, MCP servers, CLI tools, instruction files, and custom Agent commands — and lets you manage them safely.

Skills overview

What it does

  • Skills — scan every Agent directory and shared library, inspect sources, token cost and visibility, then install, update, uninstall, repair links or sync copies.
  • Agent Inspector — see what an Agent statically sees and why: visible Skills, duplicates, missing links, instruction files and context budget. Read-only.
  • MCP servers — inspect global and project configuration, add or edit servers, toggle them per Agent, and test whether a command or URL is reachable.
  • CLI tools — see what is installed, which package manager owns it, duplicates and available updates. Update or uninstall through the package manager.
  • Instruction files — scan AGENTS.md, CLAUDE.md, Cursor rules and Copilot instructions. Preview Markdown or raw content, edit existing files, or create new ones.
  • Commands — scan custom slash commands for Claude Code, Cursor, OpenCode, Gemini CLI and GitHub Copilot. Create or edit them through a plan and snapshot; Codex prompts are not supported by current Codex releases and use Skills instead.
  • App updates — check GitHub Releases, download a verified app archive, install it in place and restart Larder.
  • Projects — see the Skills, MCP servers and instruction files that belong to a project in one place.
  • Language — switch between English and Chinese in Settings; the app follows the system language on first launch.

Screenshots

Skill detail
Skill detail
Source, Agent visibility and the next safe action.
Review upstream additions
Review upstream additions
See new Skills before installing one or a batch.
MCP servers
MCP servers
Transport, launch command, status and Agent visibility.
CLI tools
CLI tools
Owners, PATH duplicates and available updates.
Instruction files
Instruction files
Scan, preview and edit global or project instructions.
History and rollback
History and rollback
Every write keeps a snapshot that can be restored.
Skills in dark mode
Light and dark
Light, dark or system appearance.
Settings
Settings
Language, motion, zoom and update preferences.

Safe by default

Larder reads local files before it writes anything.

  • Every write has a visible action or plan.
  • Configuration files are snapshotted before modification.
  • Failures are shown instead of silently ignored.
  • Local data stays local. No account, no file upload, no telemetry.

Build from source

Requirements:

  • macOS 13+ (current primary platform)
  • Windows 10 1809+ / 11 (build and acceptance path included; WebView2)
  • Go 1.26+
  • Node.js 24+
  • Wails v3 CLI

macOS is the current primary target. The Windows build and test path is already in the repository; more desktop platforms are planned.

git clone <repo-url>
cd larder
./scripts/run.sh

Useful scripts:

./scripts/run.sh    # build if needed, then open the app
./scripts/dev.sh    # development mode with frontend hot reload
./scripts/build.sh  # rebuild the app

There is no signed release build yet.

On macOS, a locally built app may require:

xattr -dr com.apple.quarantine /path/to/Larder.app

On Windows, an unsigned build may show a SmartScreen warning.

Status

0.0.8 — early development. The current focus is local truth, safe writes, and a clearer view of what each Agent actually sees.

License

MIT

Reviews (0)

No results found