A searchable directory of 120+ MCP servers

Every serious agent now ships with the same superpower: it can reach out of the chat window and do things — read your files, query a database, open a pull request, send an email, generate an image. The plumbing underneath almost all of it is the Model Context Protocol (MCP), an open standard introduced by Anthropic in late 2024 and since adopted by Claude, OpenAI's Codex and ChatGPT, and a long tail of editors and agent frameworks.

The problem is discovery. There are now thousands of MCP servers — official ones from Stripe, GitHub, Cloudflare, Hugging Face, and community ones for everything from Spotify to Blender — scattered across GitHub repos, vendor docs, and a handful of registries. Finding the right one, and knowing whether it costs anything, is the hard part.

So we built a searchable MCP Server Directory: a curated index of 120+ servers you can search semantically, filter by category and pricing, and link straight through to each provider.

What MCP actually is

MCP is a thin client/server protocol. The host (Claude Desktop, Claude Code, Codex, your IDE) speaks to one or more servers, each of which exposes three kinds of capability:

• Tools — functions the model can call (create_invoice, search_issues, run_sql).
• Resources — data the model can read (files, rows, documents).
• Prompts — reusable templates a server can offer the host.

The transport is either stdio (the host launches a local process and talks over stdin/stdout) or HTTP (a remote, often hosted, endpoint). That single abstraction is why one Gmail server works in Claude, Codex, Cursor, and Windsurf without per-app integration code.

Wiring one up

Most stdio servers are a one-line command. To give Claude Code access to your local Git repos and the web:

claude mcp add git -- uvx mcp-server-git
claude mcp add fetch -- uvx mcp-server-fetch

For an editor that reads a JSON config (Claude Desktop, Cursor), an entry looks like:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"]
    }
  }
}

Remote servers (Stripe, Notion, Linear, Hugging Face) are increasingly served over HTTP with OAuth, so you paste a URL and authenticate in the browser — no local install at all.

How we organized the directory

Each server in the index carries the metadata that actually matters when you're choosing one:

• Category — AI & LLM, Databases, Dev Tools, Communication, Search & Web, Design, Finance, and more.
• Pricing — is the underlying service free, freemium, paid, or open-source? This is the question every list omits and everyone asks first.
• Official vs community — whether the vendor maintains the server themselves.
• Transport — stdio, HTTP, or both.
• Links — homepage, source repo, and docs, plus a copy-paste install command where one exists.

A few worth knowing about:

• [Hugging Face](https://huggingface.co/settings/mcp) turns model search, datasets, and any Gradio Space into callable tools — thousands of inference demos become agent tools for free.
• [fal](https://fal.ai) and [Replicate](https://replicate.com) expose fast generative image, video, and audio models.
• [GitHub](https://github.com/github/github-mcp-server), [Stripe](https://docs.stripe.com/mcp), [Cloudflare](https://developers.cloudflare.com/agents/model-context-protocol), [Canva](https://www.canva.dev/docs/connect) ship first-party servers.
• Aggregators like [Zapier](https://mcp.zapier.com), [Composio](https://composio.dev), and [Pipedream](https://pipedream.com/docs/connect/mcp) put thousands of authenticated apps behind a single MCP endpoint.
• The [official MCP registry](https://registry.modelcontextprotocol.io) and directories like [Smithery](https://smithery.ai) and [Glama](https://glama.ai/mcp/servers) host and resolve servers by name.

Search that understands intent

The directory isn't a static table. Search runs over a local embedding index, so "send a text message" surfaces Twilio and "remember things across sessions" surfaces the Memory and vector-database servers — even when the words don't match. When the embedding model isn't available it falls back to ranked keyword search, so it always answers. It's the same in-process search stack that powers our skills library and art search.

Try it

Browse and search the full index at [/mcp-servers](/mcp-servers). If you maintain a server that should be listed, the directory is curated — tell us and we'll add it. And if you want somewhere to run agents that use these servers, that's what the rest of app.nz is for: the cloud coding agent, an OpenAI-compatible gateway, and instant deploys.