Agent Shelf

hub

MCP Server

Connect your AI tools to AgentShelf via the Model Context Protocol. Search, discover, and download agents directly from your AI assistant.

Overview

The AgentShelf MCP server lets any MCP-compatible AI tool search and discover agents from the registry. Instead of switching to the browser, your AI assistant can find the right agent for the task at hand, read its documentation, and download it — all within your coding environment.

Find agents by keyword, category, or tag

visibility

Discover

Browse featured and popular agents

download

Download

Get agent markdown to use locally

Remote Access

bolt

No installation required

AgentShelf hosts the MCP server for you. Just add the URL to your AI tool's config — no download, no build step.

Add the following to your AI tool's MCP configuration to connect directly to the hosted AgentShelf MCP server:

{
  "mcpServers": {
    "agentshelf": {
      "type": "streamable-http",
      "url": "https://your-agentshelf-url.com/api/mcp"
    }
  }
}

Claude Code / Claude Desktop

Add to .mcp.json or claude_desktop_config.json

Cursor / Windsurf / Others

Add to your tool's MCP config file. See the Setup by Tool section below.

Remote access uses the Streamable HTTP transport. All tools are read-only — no authentication required for search and discovery.

Local Installation

Download the MCP server, install dependencies, and build it.

downloadDownload MCP Server (.zip)

Install and build:

# Unzip and install
unzip agentshelf-mcp-server.zip
cd agentshelf-mcp-server
npm install
npm run build

Requirements: Node.js 18+ and npm. The server communicates over stdio — no ports or network configuration needed.

Setup by Tool

Add the MCP server to your AI tool's configuration file. Pick your tool below:

Config file:.mcp.json

Location:Project root or ~/

{
  "mcpServers": {
    "agentshelf": {
      "type": "streamable-http",
      "url": "https://your-agentshelf-url.com/api/mcp"
    }
  }
}

Important notes:

arrow_forwardReplace the path in args with the actual location where you unzipped the server
arrow_forwardSet AGENTSHELF_URL to your AgentShelf instance URL
arrow_forwardRestart your AI tool after updating the config

Available Tools

Search Agents

search_agents

Search the registry by keyword, category, or tag. Supports sorting by downloads, likes, name, or date.

Parameter	Type	Description
query	string	Free-text search
category	string	Filter by category
tag	string	Filter by tag
sort	string	downloads \| likes \| name \| createdAt
limit	number	Max results (1-50)

info

Get Agent Details

get_agent

Get full details about a specific agent including markdown content, version history, and metadata.

Parameter	Type	Description
username	string	Author's GitHub username
name	string	Agent slug ID

List Categories

list_categories

List all valid agent categories with descriptions. Useful for filtering searches.

star

Get Featured Agents

get_featured

Get the top featured agents, sorted by popularity.

download

Download Agent

download_agent

Download the raw markdown content of an agent to use locally.

Parameter	Type	Description
username	string	Author's GitHub username
name	string	Agent slug ID

Examples

Once connected, your AI assistant can use AgentShelf tools automatically. Here are some things you can ask:

chat

“Find me a code review agent”

search_agents

Searches for agents matching "code review" and shows results with downloads, likes, and descriptions.

chat

“What agents are available for DevOps automation?”

search_agents

Filters by the devops category and returns matching agents.

chat

“Show me the most popular agents”

get_featured

Returns the top featured agents sorted by popularity.

chat

“Download the code-reviewer agent by alice”

download_agent

Downloads the full markdown with frontmatter, ready to use locally.

chat

“What categories of agents are there?”

list_categories

Lists all 14 categories with descriptions to help you find what you need.

Troubleshooting

Issue	Fix
Server not showing in tool	Check the path in `args` points to `build/index.js`. Make sure you ran `npm run build` first. Restart your AI tool.
Connection errors	Verify `AGENTSHELF_URL` is correct and the AgentShelf instance is running.
No results from search	Check that agents have been published on the registry. Try a broader search or use `get_featured` to see if any agents exist.
Node.js version error	Requires Node.js 18 or higher. Check with `node --version`.