The cross-LLM marketplace for real, tool-using AI agents. Find one that fits your work, connect it in a few clicks, and put it to work in Claude, ChatGPT, Gemini, or Cursor. Built for everyone — no code required.

Real agents, or just prompts?

Real agents. They use tools and take actions to get work done — not prompt snippets you paste in. Connect one and it starts working right away.

Which AI apps can I use them in?

Claude, ChatGPT, Gemini, Cursor, VS Code — anywhere MCP works. The same agent moves between apps with nothing to rewrite.

Is it safe to connect an agent?

Security is the whole point. Most agents ship no code that runs on your machine, your API keys are sent only to the hosts an agent declares, and every listing is automatically scanned and human-reviewed before it goes live.

How do I get started?

FindAgent is in open beta — browse, search, and connect agents free today, no invite or waitlist. Creators can publish through a guided flow and keep the majority of every sale when payments launch.

What is an MCP server marketplace?

An MCP server marketplace is a catalogue where you discover, compare, and connect Model Context Protocol servers — the standard way to give an AI client tools and data. Instead of hunting through code repositories, you browse listings by use case, see how each server connects, and add it to Claude, ChatGPT, or Cursor. FindAgent is a marketplace that hosts servers, meters usage, and pays creators.

What is the difference between an MCP marketplace and an MCP registry?

A registry is a flat index of server names and repository links — you still clone, build, and run each server yourself. A marketplace adds a product layer on top: curated listings, a hosted connection option so nothing runs on your machine, usage metering, reviews, and creator payouts. On FindAgent a hosted MCP server connects over a single URL with sign-in, so you skip the local install entirely.

How do I find an MCP server?

Browse an MCP server marketplace by category or use case — databases, analytics, Shopify, GitHub, search — read the listing to see what tools the server exposes and whether it connects hosted or locally, then connect it to your AI client. On FindAgent the directory at /mcp is filterable by connection type (hosted or local) so you can pick the setup that fits your workflow.

How does an MCP server connect to Claude, ChatGPT, or Cursor?

A hosted MCP server connects over a remote URL that your AI client points at once; the marketplace runs the server and you sign in. A local server runs on your own machine over stdio. Claude, ChatGPT, and Cursor all speak the Model Context Protocol, so the same server works across them — you connect once and use it in whichever client you already have open.

Are MCP servers free?

Many MCP servers are free to connect. A marketplace can also list paid servers — one-time or subscription — where the creator sets the price and the marketplace handles billing. FindAgent is creator-first: creators keep the majority of every sale, and hosted usage is metered so paid servers bill for what you actually run.

What can MCP servers do?

An MCP server exposes tools and data to an AI client: querying a database, reading and writing files, calling an API, searching the web, or running a full workflow. Because the AI client can call those tools directly, a connected server turns a chat assistant into something that takes real actions — reconciling data, managing a store, or pulling analytics — rather than just describing how.

MCP server marketplace

An MCP server marketplace is a directory where you find, compare, and connect Model Context Protocol servers to your AI client — Claude, ChatGPT, or Cursor — instead of building and hosting each one yourself. You browse listings by use case, see how each server connects, and add it in a few steps.

A marketplace such as FindAgent goes further than a raw registry: it hosts servers so nothing runs on your machine, meters usage, and pays the people who build them. Browse the live directory of MCP servers on FindAgent to see what is available today.

What is an MCP server?

An MCP server is a program that exposes tools and data to an AI client over the Model Context Protocol — an open standard, originally from Anthropic, that Claude, ChatGPT, and Cursor all support. Where a plain chat assistant can only talk, a connected MCP server lets the assistant act: query a database, read and write files, call an API, search the web, or run a multi-step workflow. The server declares which tools it offers; the AI client calls them when a task needs them.

How to find and choose an MCP server

Finding a server is the easy part; choosing the right one is where a marketplace earns its place. When you evaluate a listing, look at four things:

Use case — what job does it do? Browse by category (databases, analytics, ecommerce, GitHub) rather than by server name.
Tools exposed — read the listing to see the exact tools the server offers, and whether they read data or also write and delete.
Connection type — hosted (connect over a URL, nothing local) or local (runs on your own machine over stdio). FindAgent labels every listing so you can filter.
Trust signals — reviews, whether the source is public, and whether the server was reviewed before listing.

Start from the MCP directory, or browse agents if you want a packaged, ready-to-run capability rather than a raw server.

Marketplace vs registry: why the difference matters

A registry lists servers; a marketplace makes them usable. The table below shows what a marketplace adds on top of a plain index.

Comparison of an MCP server marketplace and a raw MCP registry across listings, hosting, discovery, billing, and creator economics.
Dimension	MCP marketplace	Raw registry
What you get	Curated, described listings with reviews	A flat index of names + repo links
Running the server	Hosted option — connect over a URL, nothing local	Clone, build, and run it yourself
Discovery	Browse by use case, filter by connection type	Search by name, no use-case grouping
Usage + billing	Metered hosted usage; paid servers supported	None — you self-host
Creator economics	Creator-first payout on paid servers	No monetisation layer

How connecting works across Claude, ChatGPT, and Cursor

Because the Model Context Protocol is a shared standard, one server works across every client that speaks it. There are two connection paths:

Hosted — the marketplace runs the server. Your AI client points at a single remote URL, you sign in once, and the server is available with no install. This is the path that reaches ChatGPT and mobile clients, and the only one where usage is metered.
Local — the server runs on your own machine over stdio, useful for offline or development work. You add it to your client's config and it runs where your files are.

Either way you connect once and use the server in whichever client you already have open — install once, run anywhere.

Frequently asked questions

What is an MCP server marketplace?: An MCP server marketplace is a catalogue where you discover, compare, and connect Model Context Protocol servers — the standard way to give an AI client tools and data. Instead of hunting through code repositories, you browse listings by use case, see how each server connects, and add it to Claude, ChatGPT, or Cursor. FindAgent is a marketplace that hosts servers, meters usage, and pays creators.
What is the difference between an MCP marketplace and an MCP registry?: A registry is a flat index of server names and repository links — you still clone, build, and run each server yourself. A marketplace adds a product layer on top: curated listings, a hosted connection option so nothing runs on your machine, usage metering, reviews, and creator payouts. On FindAgent a hosted MCP server connects over a single URL with sign-in, so you skip the local install entirely.
How do I find an MCP server?: Browse an MCP server marketplace by category or use case — databases, analytics, Shopify, GitHub, search — read the listing to see what tools the server exposes and whether it connects hosted or locally, then connect it to your AI client. On FindAgent the directory at /mcp is filterable by connection type (hosted or local) so you can pick the setup that fits your workflow.
How does an MCP server connect to Claude, ChatGPT, or Cursor?: A hosted MCP server connects over a remote URL that your AI client points at once; the marketplace runs the server and you sign in. A local server runs on your own machine over stdio. Claude, ChatGPT, and Cursor all speak the Model Context Protocol, so the same server works across them — you connect once and use it in whichever client you already have open.
Are MCP servers free?: Many MCP servers are free to connect. A marketplace can also list paid servers — one-time or subscription — where the creator sets the price and the marketplace handles billing. FindAgent is creator-first: creators keep the majority of every sale, and hosted usage is metered so paid servers bill for what you actually run.
What can MCP servers do?: An MCP server exposes tools and data to an AI client: querying a database, reading and writing files, calling an API, searching the web, or running a full workflow. Because the AI client can call those tools directly, a connected server turns a chat assistant into something that takes real actions — reconciling data, managing a store, or pulling analytics — rather than just describing how.

FindAgent is in open beta — anyone can connect now. Browse the marketplace to get started, or browse the MCP server directory now.