A Model Context Protocol server that gives any AI assistant browse and discovery of the three.ws marketplace over stdio. Search and page the published agents, drill into one by id, and explore the reusable skills catalog — all from the public three.ws API.

No API key, no signer, no payment — every call hits the public /api/marketplace and /api/skills endpoints. Creating or publishing agents and skills is the authenticated write path on the HTTP API; this server exposes browse/discovery only.

Install

npm install @three-ws/marketplace-mcp

Or run with npx (no install):

npx @three-ws/marketplace-mcp

Quick start

Claude Code, one line:

claude mcp add marketplace -- npx -y @three-ws/marketplace-mcp

Claude Desktop / Cursor (claude_desktop_config.json or mcp.json):

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

Inspect the surface with the MCP Inspector:

npx -y @modelcontextprotocol/inspector npx @three-ws/marketplace-mcp

Tools

| Tool | Type | What it does | | ------------------- | --------- | ---------------------------------------------------------------------------------------------------- | | browse_agents | read-only | Search & page the agent marketplace — filter by category and query, sort, paginate by cursor. | | agent_detail | read-only | Fetch one published agent by id (system prompt, capabilities, author, ratings, skill prices, avatar). | | agent_categories | read-only | List agent categories with the count of published agents in each, plus the overall total. | | browse_skills | read-only | Search & page the skills catalog — filter by query and category, sort, paginate by cursor. | | skill_categories | read-only | List skill categories that have at least one public skill, each with a slug, label, and count. |

Input parameters

browse_agents — category (optional slug), q (optional free text), sort (recommended | recent | popular | top_rated, default recommended), limit (1–48, default 24), cursor (optional pagination cursor).

agent_detail — id (required: the agent's UUID, from a browse_agents card).

agent_categories — no parameters.

browse_skills — q (optional free text), category (optional slug), sort (popular | new | az, default popular), limit (1–50, default 20), cursor (optional pagination cursor).

skill_categories — no parameters.

Example

// browse_agents
> { "q": "code review", "sort": "top_rated", "limit": 3 }
{
  "ok": true,
  "items": [
    {
      "id": "a1b2c3d4-…",
      "name": "Code Reviewer",
      "description": "Reviews diffs for bugs and style.",
      "category": "programming",
      "tags": ["code", "review"],
      "rating_avg": 4.8,
      "rating_count": 22,
      "views_count": 1340,
      "forks_count": 41,
      "thumbnail_url": "https://…",
      "avatar_glb_url": "https://…"
    }
  ],
  "next_cursor": "3"
}

Pass next_cursor back as cursor to fetch the following page.

Requirements

• Node.js >= 20.
• Network access to https://three.ws (or your own THREE_WS_BASE).

Environment variables

| Variable | Required | Default | | --------------------- | -------- | ------------------ | | THREE_WS_BASE | no | https://three.ws | | THREE_WS_TIMEOUT_MS | no | 20000 |

Links

• Homepage: https://three.ws
• Changelog: https://three.ws/changelog
• Issues: https://github.com/nirholas/three.ws/issues
• License: Apache-2.0 — see LICENSE