Versuz

by TomaTV

1 stars
267 downloads
Not rated
GitHub Website

About

Versuz is an MCP (Model Context Protocol) server that connects Claude Code to the open Versuz marketplace. It lets you browse, inspect, and install skills and CLAUDE.md files as native tools directly from your conversation, without leaving the chat.

Details

Author
TomaTV
GitHub stars
1
Downloads
267
Categories
Other

- Full-text search across skills and CLAUDE.md (versuz_search)
- Paginated browse with category, tier, and sort filters (versuz_list_skills)
- List CLAUDE.md files by category (versuz_list_claude_md)
- Retrieve full details for a slug, including Elo, prior, and GitHub URL (versuz_get)
- Install free skills or CLAUDE.md files directly to your project (versuz_install)

Setting up with Highlight

This MCP is not yet compatible with Highlight’s one-click setup. However, you can still use it with Highlight by following these steps:

  1. Download and install Highlight from highlightai.com/download
  2. Navigate to the plugins tab and select "Add Custom Plugin"
  3. Configure the plugin with the settings below
    Plugin Name Versuz
    Command (node, npx, python, etc.)

    Please refer to the README for specific instructions on how to obtain API keys or other required environment variables.

  4. Enable "Start Automatically" if you want the plugin to start when Highlight launches

From the repository

Install the server with claude mcp add versuz npx -y @versuz/mcp or add it manually to your project’s .mcp.json configuration. After restarting Claude Code, five Versuz tools appear in the tool list and can be invoked by name.

versuz_search

Search the Versuz marketplace for Claude skills and CLAUDE.md files. Returns up to 15 matches per kind, ranked by prior + stars. Use when the user asks to find a skill for a task (e.g. 'PDF generation', 'database migration') or a CLAUDE.md for a project type.

versuz_list_skills

List Claude skills from the marketplace. Filter by category, tier, or sort order. Use to browse what's available before installing.

versuz_list_claude_md

List CLAUDE.md project-context files from the marketplace. Filter by category (nextjs, react, python-data, backend-api, mobile, devops, ml-training, generic).

versuz_get

Get full details for a specific skill or CLAUDE.md by slug — Elo, prior, license, GitHub link, bundle files, etc. Useful before deciding to install.

versuz_install

Install a skill (writes to .claude/skills/<slug>/SKILL.md) or a CLAUDE.md (writes to ./CLAUDE.md at project root). Only free items are installable via MCP — premium items return a buy URL the user must visit first.

versuz_battle

Head-to-head comparison of two skills or CLAUDE.md files. Returns rank, bench score, prior, stars, tier, and a verdict (winner + delta). Use when the user asks to compare two items, pick a winner, or decide between options. Examples : 'compare pdf-generator vs anthropic-pdf', 'which is better : nextjs-supabase or nextjs-prisma'.

versuz_submit

Submit a SKILL.md or CLAUDE.md from a public GitHub repo to the Versuz registry. Requires a GitHub Personal Access Token (PAT) configured via VERSUZ_GITHUB_TOKEN env var — the same token format as `npx versuz login`. Only the repo owner or a verified org member can submit. Free tier only via MCP ; premium submissions go through the web at versuz.dev/submit. Use when the user asks to publish their skill on Versuz, share a SKILL.md they wrote, or list their repo.

Claude Desktop / Cursor

Paste into your MCP client config file to install this server.

{
    "mcpServers": {
        "versuz": {
            "versuz": {
                "command": "npx",
                "args": [
                    "-y",
                    "@versuz/mcp"
                ]
            }
        }
    }
}

McpServers

{
    "versuz": {
        "command": "npx",
        "args": [
            "-y",
            "@versuz/mcp"
        ]
    }
}

@versuz/mcp

The MCP server for the Versuz marketplace.

Plug Claude Code into the open Versuz registry. Browse, inspect and install skills + CLAUDE.md as native tools, without leaving your conversation.

Beta · MIT · Source

</div>

---

What it does

Model Context Protocol is Anthropic's standard for letting Claude Code call external tools. @versuz/mcp exposes the Versuz catalog as 5 tools, so you can say things like :

> "find me a SQL migration skill and install it"

> "list the top 10 CLAUDE.md files for Next.js projects"

> "show me the prior score and Elo for the pdf-pro skill"

Claude calls the tools, summarises results, downloads the right files. No copy/paste, no browser switch.

Install

claude mcp add versuz npx -y @versuz/mcp

Or manual via .mcp.json in your project :

{
  "mcpServers": {
    "versuz": {
      "command": "npx",
      "args": ["-y", "@versuz/mcp"],
      "env": {
        "VERSUZ_API": "https://versuz.dev"
      }
    }
  }
}

Restart Claude Code. The 5 Versuz tools appear in the tool list.

Tools exposed

| Tool | Description |
|---|---|
| versuz_search | Full-text search across skills + CLAUDE.md. Returns top 15 of each kind. |
| versuz_list_skills | Paginated browse with category / tier / sort filters. |
| versuz_list_claude_md | Same shape for CLAUDE.md (categories : nextjs · react · python-data · backend-api · mobile · devops · ml-training · generic). |
| versuz_get | Full details for a slug — Elo, prior, license, GitHub URL, bundle files, tools, model. |
| versuz_install | Download to .claude/skills/<slug>/SKILL.md (skill) or ./CLAUDE.md (claude_md). Free items only. Premium returns a buy_url. |

Example session

You : "I need a skill that can extract tables from PDFs."

Claude :

[invoking versuz_search({query: "pdf table extraction"})]

Found 3 candidates :
- pdf-table-extractor (prior 1721, ★ 8.2k, FEATURED $4.99)
- pdf-extract-anthropic (prior 1846, ★ 12.4k, FREE)
- table-from-pdf-claude (prior 1432, ★ 950, FREE)

Recommend pdf-extract-anthropic — highest prior + free. Install ?

You : "yes"

Claude :

[invoking versuz_install({slug: "pdf-extract-anthropic"})]

✓ Wrote .claude/skills/pdf-extract-anthropic/SKILL.md (3.2 KB)
Source : github.com/anthropics/skills/blob/main/pdf-extract/SKILL.md

Config

| Env var | Default | Effect |
|---|---|---|
| VERSUZ_API | https://versuz.dev | API host. Set to http://localhost:3000 for local dev. |

You can also pass via the MCP config :

{
  "mcpServers": {
    "versuz": {
      "command": "npx",
      "args": ["-y", "@versuz/mcp"],
      "env": { "VERSUZ_API": "http://localhost:3000" }
    }
  }
}

Install paths

Tools write files relative to the project root where Claude Code is running, not where the MCP server is :

| Kind | Destination |
|---|---|
| skill | <project>/.claude/skills/<slug>/SKILL.md |
| claude_md | <project>/CLAUDE.md |

Override via the cwd argument on versuz_install if needed.

Premium handling

Premium items return 402 Payment Required from the API. The MCP server forwards a clear message :

Error: Premium skill — purchase required to download via MCP.
Buy at: https://versuz.dev/buy/skill/<slug>

Once you've purchased on the web, future installs will work (in v0.2 — currently the MCP doesn't pass user auth, only free items are downloadable).

Bundle support (v0.2)

Same caveat as the CLI : bundled skills (SKILL.md + scripts/refs) download only the SKILL.md for now. A bundle: true arg on versuz_install is on the roadmap once /api/v1/skills/<slug>/bundle.zip ships.

Roadmap (v0.2)

- [ ] versuz_submit tool — publish a skill from your own GitHub repo (mirrors CLI)
- [ ] GitHub Device Flow auth — for premium downloads via MCP
- [ ] Bundle download — zip support
- [ ] versuz_judge — kick a quality-judge run from MCP (admin only)

Development

cd mcp-server
npm install
node bin/server.js          # runs on stdio, waits for MCP client

The server logs [versuz-mcp] running on stdio · api=<url> to stderr on startup. stdout is reserved for the MCP protocol.

Test it standalone with @modelcontextprotocol/inspector :

npx @modelcontextprotocol/inspector node bin/server.js

Stack

- @modelcontextprotocol/sdk ^1.0
- Stdio transport (no network listener)
- Pure Node 18+, ESM

License

MIT — see LICENSE.

No reviews yet — be the first

Sign in to leave a review

Use Google, GitHub, or an email account so ratings stay tied to real people.

Email sign in

No reviews posted yet.