# LoopSkill [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/wisechef-ai/loopskill-api  
**GitHub Stars:** 2  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/loopskill

## Description
Search, install and compose AI agent skills, loops and bundles from the LoopSkill marketplace.

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What LoopSkill MCP server does

The LoopSkill MCP server combines a local command-line workflow with an optional self-hosted registry. The CLI works with skills already stored on a machine and can create a lockfile snapshot of those installations. It can then compare the snapshot with the current filesystem to identify changed or missing skills.

The registry adds a catalog for skills, bundles, and loops. Its documented API includes full-text and vector-hybrid skill search, skill details with versions, public bundle discovery, loop listing, and a health endpoint. The registry also exposes a synchronous endpoint for running a loop and returning its check result.

This setup separates local inspection from network access. You can use import and diff without an account, registry, or network connection, then opt into registry-backed operations when you need to pull or apply content.

## How it works

The LoopSkill MCP server scans skill directories associated with supported agent environments and records the result in a JSON lockfile. A later diff compares that lockfile with a live scan. Modified skills and skills present on only one side are reported, and drift produces a nonzero exit code suitable for scripts or CI. A clean comparison returns success.

The network-capable path is isolated in the pull module. Import and diff do not load that module, and the project includes a test that blocks socket creation while running those commands. Pull and apply are the commands that contact a registry or make changes based on registry content.

For scheduled loops, the repository has a narrower implementation. Loop execution can be invoked synchronously through the API, while cron materialization currently targets Hermes through its jobs file. Loop telemetry is emitted only when the loop prompt invokes the supplied emission script.

## Setup and configuration

Install the CLI from the repository into a project-local Python virtual environment, then invoke it from that environment. The repository documents a drift demonstration using fabricated skill directories, a generated lockfile, and a subsequent live comparison. Replace the demo home path with the location containing the skills you want to inspect.

The optional registry can be started with Docker Compose. The default stack uses SQLite, creates development secrets, seeds a starter catalog, and prints a development API key at startup. Production deployments can override the key with `WR_API_KEY`; the repository also documents a PostgreSQL and production configuration path.

## Tools and capabilities

The LoopSkill MCP server supports these documented workflows:

- Search skills in the registry.
- Inspect skill details and available versions.
- Discover public bundles.
- List loops and their safety contracts.
- Run a loop synchronously through the API.
- Import installed skills into a lockfile.
- Diff a lockfile against the live filesystem.
- Pull or apply registry content when network access is explicitly used.
- Compose skills, loops, and bundles through the marketplace workflow.

The local import and diff commands are useful when network access must be avoided. Registry operations are appropriate when an agent needs catalog content or a self-hosted source of truth.

## Limitations and notes

The LoopSkill MCP server does not make import or diff network calls, but pull and apply do. Review commands before allowing them to write into real skill directories. The scheduled-loop path is not cross-vendor: cron installation is implemented for Hermes and refuses unsupported hosts rather than creating an unusable schedule.

Loop telemetry is not automatic. A loop must explicitly call the emission script from its prompt for a run to appear in telemetry. The registry's starter catalog contains ten vetted loops, but the README does not establish that every loop supports the same execution behavior or scheduling path.

_Full upstream README: https://allmcps.com/mcp/loopskill/readme_