Log inSubmit server
Claude Skills MCP logo

Claude Skills MCP

by K-Dense-AI

Searches curated Claude Agent Skills semantically and exposes progressive skill loading through MCP.

InstallRepositoryWebsite
Full README Compare alternatives Embed widget Claim this listing

Install Claude Skills MCP

Pick your client, then add it in one click or paste the config.

STDIOPython
~/Library/Application Support/Claude/claude_desktop_config.json
claude_desktop_config.json
{
  "mcpServers": {
    "k-dense-ai-claude-skills-mcp": {
      "command": "uvx",
      "args": [
        "claude-skills-mcp"
      ]
    }
  }
}

Settings → Developer → Edit Config, merge this into mcpServers, then fully quit and reopen Claude Desktop.

We couldn’t confirm it starts correctly

We ran the install command but it didn't respond within our test window — this can mean a slow first-time install rather than a real problem.

uvx claude-skills-mcp

No response to initialize.

This is an experimental automated check and can have false negatives — missing environment variables, a slow cold install, etc. It doesn’t necessarily mean something’s wrong. Last checked 1mo ago.

Overview

K-Dense-AI/claude-skills-mcp MCP server lets MCP-compatible AI clients discover and retrieve Claude Agent Skills through semantic search. A lightweight stdio frontend starts quickly, then downloads and launches a backend with vector-search dependencies on first use. It can index GitHub repositories and local skill directories, with default sources from Anthropic, K-Dense, and the user's local Claude skills path. Reach for it when an assistant needs to find specialized skills across supported sources, but note that the project is no longer hosted or maintained.

Use cases

  • Search for a skill matching an agent task
  • Load skill files progressively during a workflow
  • Index skills from GitHub repositories
  • Combine remote repositories with local Claude skills
  • Inspect the available skill inventory

Key features

  • Semantic skill search with vector embeddings
  • Progressive loading of metadata, content, and files
  • GitHub and local-directory skill sources
  • Hourly source checks and re-indexing
  • Three MCP tools for search and retrieval
  • Local operation without API keys

Security & health

What it can reach and how reliably it runs — from our own checks, not the maintainer’s claims.

ActiveLast checked 1d ago

Recent health check succeeded.

20/20 checks healthy over the last 77d

Supply-chain signal

No high-severity advisories surfaced by our automated scan.

Critical 0High 0Medium 0Low 0

Scanned 8d ago via OSV.dev · claude-skills-mcp (PyPI)

Adoption & maintenance

Factual signals from GitHub, npm, and our automated checks — not a rating.

GitHub stars409Stargazers on the source repository.
Last commit3mo agoMost recent push to the default branch.
Install checkInconclusiveDidn't respond in our test window — often a slow first install.
Directory activity5 viewsConfig copies, upvotes, and views on AllMCPs.

How Claude Skills MCP works

What K-Dense-AI/claude-skills-mcp MCP server does

K-Dense-AI/claude-skills-mcp MCP server makes Claude Agent Skills available to MCP-compatible AI applications. It indexes skill metadata and content, then uses vector embeddings and semantic similarity to find skills related to a task description. The project is intended for assistants and coding tools that need to discover relevant skills without relying on a platform-specific native skills implementation.

Default content comes from Anthropic's official skills repository, K-Dense's Scientific Agent Skills collection, and ~/.claude/skills when that local directory exists. Additional GitHub repositories and local directories can be configured as skill sources.

How it works

The distribution separates the MCP entry point from the heavier search implementation. The frontend is a small stdio package designed to start quickly, while the backend contains the vector-search stack based on PyTorch and sentence-transformers. On the first run, the frontend downloads and starts the backend in the background; later starts can reuse the downloaded components.

Search follows a progressive disclosure model. An assistant can first identify relevant skill metadata, load the complete skill content when needed, and then request individual files such as scripts, data, or reference material. The server also checks configured sources hourly and re-indexes them when changes are detected.

Setup and configuration

The final release, version 1.1.0, remains available from PyPI, and the project supports Python 3.12 and 3.13. A direct standalone launch uses uvx claude-skills-mcp. Cursor can run the same package through an MCP configuration with uvx as the command and claude-skills-mcp as its argument.

To customize the source repositories, local paths, embedding model, or content limits, generate an example configuration with uvx claude-skills-mcp --example-config, edit the resulting JSON, and start the server with --config config.json.

Tools and capabilities

The MCP interface exposes three tools:

  • find_helpful_skills searches for skills using a task description and semantic similarity.
  • read_skill_document retrieves a selected file from a skill, including scripts, data, or references.
  • list_skills returns the inventory of loaded skills for exploration or debugging.

The K-Dense-AI/claude-skills-mcp MCP server can therefore support both targeted discovery and inspection of the files behind a selected skill. Its source configuration is suitable for combining public GitHub skill repositories with directories maintained locally.

Limitations and notes

K-Dense-AI/claude-skills-mcp MCP server is no longer hosted or maintained. The repository identifies version 1.1.0 as the final release and says that major AI platforms now support Agent Skills natively. New issues and pull requests may not be reviewed, so teams extending it should consider forking the repository.

The first backend download is comparatively large and may take roughly 60–120 seconds because of the retrieval dependencies. The frontend is designed to avoid client startup timeouts, but that initial setup still requires the backend components to be downloaded. The project is licensed under Apache License 2.0.

Read the full README View source on GitHub

Reviews

No reviews yet — be the first to share how it worked for you.

Frequently Asked Questions about Claude Skills MCP

Run `uvx claude-skills-mcp`. The final release is available from PyPI, and the frontend downloads the backend on first use.

It provides `find_helpful_skills` for semantic search, `read_skill_document` for retrieving skill files, and `list_skills` for viewing the loaded inventory.

Yes. It supports local directories as skill sources and checks `~/.claude/skills` by default when that directory exists.

No. The repository states that it is no longer hosted or maintained and that version 1.1.0 is the final release.

Related MCP servers

All in Aggregators All alternatives
Add a badge to your READMEShow it’s listed — live SVG, Markdown or HTML.

AllMCPs Directory Badge

Full Badge Customizer

Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.

Badge Style:
Live Dynamic SVG PreviewClaude Skills MCP AllMCPs Directory Badge
Markdown (GitHub README)
[![AllMCPs](https://allmcps.com/api/badge/k-dense-ai-claude-skills-mcp?style=directory)](https://allmcps.com/mcp/k-dense-ai-claude-skills-mcp)
HTML Embed
<a href="https://allmcps.com/mcp/k-dense-ai-claude-skills-mcp"><img src="https://allmcps.com/api/badge/k-dense-ai-claude-skills-mcp?style=directory" alt="Claude Skills MCP on AllMCPs" /></a>