mcpub.dev: The Open, Free Directory of Remote MCP Endpoints for AI Agents | Tečaj kibernetske varnosti in izobraževanje zaposlenih po ZinfV

mcpub.dev: The Open, Free Directory of Remote MCP Endpoints for AI Agents

Tools

• MCPUB • OPENCLAW • ZEROCLAW • AI AGENTS • MCP • LLM •

Every website could be an MCP server and mcpub.dev is where they register

The Model Context Protocol (MCP) is rapidly becoming the standard way AI agents communicate with external services. As the number of MCP servers grows, there is a necessity to expose your MCP to the world. mcpub.dev is open, neutral place for agents to discover MCP services. It is a free, no-registration, no-vendor-lock directory of remote MCP endpoints, accessible by humans users and any AI agents.

What Is mcpub.dev?

mcpub.dev is a public directory of remote MCP servers. Anyone can submit an endpoint and anyone can search and connect. To list an MCP server you do not need an account or a GitHub login, and there is no review process and no fees.

The only verification: a file called /.well-known/mcp.json must exist on your domain. It can have any content or even be empty, but must be reachable from the web. That’s it, and if the file is there, your endpoint is accepted and immediately live in the directory. This design is intentional because mcpub is not a curator, but an agentic “telephone book” for MCP infrastructure.

Why This Matters: An Agentic Layer for the Web

The web has HTTP. Email has SMTP. AI agents get MCP — and now, with mcpub, they have an open discovery layer to go with it.

When an AI agent needs a capability it doesn’t have built in — a weather feed, a database query tool, a code executor, a payment API — it can query mcpub and find an actual MCP endpoint that provides it. No human in the loop, no hardcoded list of integrations, no vendor deciding what the agent is allowed to use.

This is what an agentic web looks like in practice: AI agents that can autonomously discover, connect to, and use services they’ve never seen before.

mcpub itself is an MCP server. An AI agent connects to https://mcpub.dev/mcp and gains four tools:

search — find endpoints by keyword, with pagination
list_all — browse the full directory
submit — register a new endpoint
get — look up a specific endpoint by URL

An agent can search for "calendar", get back a list of MCP servers that offer calendar capabilities, and connect to one — all without human intervention. That is the agentic layer.

How to Submit Your MCP Endpoint

Submitting to mcpub is easy and takes two steps.

Step 1: Create a file at /.well-known/mcp.json on your domain. Any content is fine:

{}

Step 2: Submit via curl:

curl -X POST https://mcpub.dev/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0", "id": 1,
    "method": "tools/call",
    "params": {
      "name": "submit",
      "arguments": {
        "url": "https://your-server.com",
        "description": "what your MCP server does"
      }
    }
  }'

Your endpoint is live immediately, there is no waiting for approval. No moderation queue. no badge, no moderator: just have a remote MCP to get agentic exposure.

How Agents Connect

Point any MCP-compatible agent at https://mcpub.dev/mcp. Here is a search query in raw JSON-RPC:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "search",
    "arguments": {
      "query": "weather",
      "limit": 10,
      "offset": 0
    }
  }
}

Results come back with total, offset, limit, and results — so agents can paginate through thousands of endpoints without any human assistance.

No Vendor Lock-In, By Design

Most AI tool directories today are closed platforms. They decide who gets listed and sometimes charge for visibility. They own the relationship between your tool and the agents that use it.

mcpub is the opposite because it’s source code is open. The API is standard JSON-RPC over HTTT, there is no account, no API key, no SDK to install. If mcpub.dev ever goes away, the protocol and the directory format are simple enough that anyone can run their own instance. It is just your MCP endpoint, your domain, your description and the directory just points to your server.

Self-Hosting Your Own MCP Server with suckless-mcp

Want to expose your own CLI tools to AI agents without writing a framework? suckless-mcp is a lightweight, self-hostable MCP server written in Rust. It has only one binary, one directory of skills, and no boilerplate or dependencies. It turns any command-line tool into a compliant MCP endpoint. Drop a skill definition into a directory, start the binary, and your tool is immediately reachable by any MCP-compatible agent from your domain name.

Install it in one line:

curl -fsSL https://raw.githubusercontent.com/roverbird/suckless-mcp/main/install.sh | sh

Also available on crates.io.

Once your self-hosted MCP server is running, submit it to mcpub — and it joins a growing, open, agent-searchable web of MCP endpoints.

The Bigger Picture

MCP is to AI agents what HTTP was to browsers: a common protocol that makes the ecosystem composable. But protocols only reach their potential when there is open infrastructure around them.

mcpub is a small piece of that infrastructure. Free to use, free to ignore, free to fork. The goal is a web where any AI agent can find any capability it needs — without a gatekeeper deciding what it’s allowed to discover.

4,700+ endpoints and counting. No gatekeepers, just endpoints for all.

🌐 mcpub.dev · 🤖 https://mcpub.dev/mcp · GitHub

*Questions? Open an issue on the suckless-mcp GitHub repo or get in touch.

Informacijski pooblaščenec in URSIV preverjata dokumentacijo, ne le sisteme. Ste pripravljeni?

ZInfV-1 zahteva dokazljivo usposabljanje zaposlenih — evidence udeležbe so med prvimi dokumenti, ki jih preveri inšpekcija. Naš praktičen tečaj (prilagojen vaši organizaciji) pokrije zakonsko obveznost in zgradi varnostno kulturo v enem koraku. Pridobite ponudbo za vašo organizacijo →

clanek065En