M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
A

Agent Skills MCP

Community
MCP server that exposes the Agent Skills catalog to AI coding agents via progressive disclosure
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #133 of 438
Stars
★ 7.1k Very popular
Transport
stdio (local process)
Runtime
Node.js >=22
Credentials
No credential needed
License
Other / unspecified
Last commit
Tools
4
59FMRS · C

agent-skills-mcp is the MCP server from the community project tech-leads-club/agent-skills, exposing its skills catalog through four tools — list_skills, search_skills, read_skill, and fetch_skill_files — using progressive disclosure so any MCP-compatible client can search and load skill content on demand. It is not an officially maintained protocol or vendor server, and users should independently verify the licensing and security-scanning claims for any third-party skills they consume.

Strongest · Maintenance 15/20 Weakest · Reliability 6/20

Reliability
6/20
Security and permissions
12/20
Maintenance
15/20
Documentation
13/20
Setup experience
13/20
Why each score
Reliability 6/20
No server manifest was cached, and no actual CI workflow file contents or committed test files were supplied as evidence — only indirect README signals like a build-status badge, semantic-release, and Nx Cloud. The four MCP tools (list_skills, search_skills, read_skill, fetch_skill_files) are described in a single line each, so there is no way to verify that declared tool schemas match real behavior, nor whether errors/dependencies are controllable. Per the static calibration rule, without verifiable execution evidence the score cannot exceed 12; a 6 reflects a plausible happy path but no reviewable proof.
Security and permissions 12/20
The four MCP tools appear, from their descriptions, to be read-only catalog browsing/search/fetch operations with no evident dangerous side effects (no delete/pay/remote-exec). The README claims defense-in-depth (sanitization, path isolation, symlink guards, atomic lockfile, audit trail) and Snyk Agent Scan scanning, but these claims are attributed to the CLI install flow rather than the MCP server itself, and the referenced SECURITY.md/threat-model content was not included in the evidence, so implementation details can't be verified. Network boundaries for CDN fetches and local cache write behavior are not clearly disclosed. No red-line risk is evident (no credential theft or covert exfiltration found), but incomplete scoping/data-flow disclosure warrants a deduction to 12.
Maintenance 15/20
The repo shows 4988 stars, only 8 open issues, is not archived, uses semantic-release for publishing, has npm version/download badges, and has both a CONTRIBUTING.md and a referenced SECURITY.md (vulnerability reporting channel) — signs of active governance. However, the repository-level license metadata is 'NOASSERTION', which doesn't cleanly match the README's claimed dual MIT (engine code) / CC-BY-4.0 (skill content) licensing, introducing licensing ambiguity. This nets a moderate deduction, landing at 15.
Documentation 13/20
The README provides fairly thorough coverage of the overall project, CLI usage, and installation, with a dedicated MCP server section including a tool table and a JSON install snippet, and it points to packages/mcp/README.md for full client setup and an error reference — but that sub-document's actual content was not provided in the evidence, so tool parameters, response formats, limits, timeouts, and cost/troubleshooting details cannot be verified as truly complete. Layered documentation links exist but key sub-document content is missing, so the score is reduced to 13.
Setup experience 13/20
The README supplies a concise, standard MCP client JSON install example (npx -y @tech-leads-club/agent-skills-mcp) that matches common MCP configuration conventions, appearing simple and low-friction. Per the static calibration rule, without actual execution evidence (no CI/test content proving the npx command successfully starts and completes the MCP init/tool-list handshake), the score cannot be rated highly; 13 reflects 'plausibly works but unverified/manual confirmation needed.'

Static review · not runListed 2026-08-06

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

  • Developers already using the tech-leads-club/agent-skills catalog who want to search and invoke skills directly from an MCP-enabled coding assistant
  • Workflows that benefit from progressive, on-demand loading of skill content rather than pulling the whole catalog at once

Not for

  • Teams requiring an MCP implementation maintained directly by an official product or protocol owner (this is a community project from Tech Leads Club, not an Anthropic or other official offering)
  • Use cases needing the server itself to execute code, access the filesystem, or call external cloud services (the README does not document such tools)

Required permissions

  • Running via npx requires network access to download and execute the @tech-leads-club/agent-skills-mcp package
  • The MCP server needs access to the skills catalog data (via CDN or local cache) to support listing, searching, and reading

Risks and side effects

  • The README notes that roughly 13% of skills in open marketplaces contain critical vulnerabilities; while this project claims static analysis and scanning of its skills, running third-party packages via npx -y still requires trusting the published package
  • The repository's license field is NOASSERTION, and some skills come from community or third-party authors retaining their own original licenses — verify individual skill licensing before commercial or compliance-sensitive use
  • The README does not provide detailed input/output parameters or error codes for the tools, so full integration requires consulting packages/mcp/README.md

Setup

Before you start

Runtime：Node.js >=22

Add an mcpServers entry named agent-skills to your MCP-compatible client's configuration (e.g. Claude Code, Cursor, VS Code), setting command to npx and args to ["-y", "@tech-leads-club/agent-skills-mcp"], then save and restart the client; npx will fetch and run the server on demand without requiring a global install.

.mcp.json 
{"mcpServers":{"agent-skills":{"command":"npx","args":["-y","@tech-leads-club/agent-skills-mcp"]}}}

Shown for Claude Code. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "agent-skills": {
      "command": "npx",
      "args": [
        "-y",
        "@tech-leads-club/agent-skills-mcp"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add agent-skills -- npx -y @tech-leads-club/agent-skills-mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

Confirm that list_skills, search_skills, read_skill, and fetch_skill_files appear in the client's MCP tool list, then ask the client to list all available skills; a returned catalog proves the connection works.

Troubleshooting

  1. If the client fails to start the server, confirm the local Node.js version is >=22 as indicated by the repository badges
  2. If the first run is slow or requires connectivity, confirm network access to the npm registry so npx can fetch the package on demand
  3. For issues related to catalog or cache staleness, note the CLI-side cache defaults to ~/.cache/agent-skills/ and can be cleared and retried
  4. Consult packages/mcp/README.md in the repository for full client setup, caching behavior, and an error reference

Things to try

Once connected, you can ask your AI assistant things like:

  • List all available skills in the agent-skills catalog
  • Search for a skill that does browser automation
  • Read the main instructions for the tlc-spec-driven skill
  • Fetch the reference files for the playwright-skill skill

Tools 4

list_skills read-only
Browse the full skills catalog by category (should only be called when the user explicitly asks to browse/list the catalog)
search_skills read-only
Find skills by intent using fuzzy search
read_skill read-only
Load a skill's main SKILL.md instructions
fetch_skill_files read-only
Fetch specific reference or template files for a skill on demand

Use cases

Search and browse the Agent Skills catalog directly from within a coding assistant instead of switching to a terminal to run the CLI
Load a specific skill's detailed instructions and reference files on demand, avoiding injecting the entire catalog into context at once
Dynamically extend Claude Code, Cursor, and similar clients with domain-specific workflows (cloud architecture, browser automation, design-to-code, security review, etc.)

Supported clients

Claude Code
Cursor
VS CodePartial support

Listed from the project's documentation, not tested by this site.

Overview

agent-skills-mcp is part of the tech-leads-club/agent-skills project, a secure, validated skill registry for professional AI coding agents. This MCP server exposes that skills catalog to MCP-compatible clients using progressive disclosure: clients search or browse first, then fetch only the specific skill instructions or reference files they need, rather than loading the whole catalog at once. The underlying catalog is maintained by the project's CLI (@tech-leads-club/agent-skills), and skills are scanned with tools like Snyk Agent Scan before publishing. The repository's overall license field is NOASSERTION, but the README states the CLI/engine code is MIT-licensed, maintainer-authored skill files are CC-BY-4.0, and third-party skills retain their original licenses.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 64.1k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 1da2195067c1 Data synced 2026-11-04 Read the FMRS scoring method