Every published agent is reachable over MCP — prompt recipes, doer agents, skills bundles, code agents, and departments alike. Open an agent and follow its steps: install it locally with the CLI, or add its hosted MCP server URL as a connector. It then runs inside Claude, ChatGPT, Gemini, Cursor, or any MCP client.

Browsing the catalogue with pricing from inside your AI client is rolling out — on clients that support it, it appears as a store panel right in the chat, not a separate website.

You don't even have to pick one first: connect the FindAgent platform MCP and your AI can find and connect the right agent for you mid-conversation.

Don't want to fill out submission forms by hand? Add the FindAgent platform MCP to your AI client and let your own assistant do it: it reads your repo and writes the listing for you — you just review and submit. Your assistant does the work, so it doesn't cost you (or us) a separate AI bill.

claude mcp add --transport http findagent https://mcp.findagent.cloud/mcp

{
  "mcpServers": {
    "findagent": {
      "url": "https://mcp.findagent.cloud/mcp"
    }
  }
}

A few of the tools it exposes

• findagent_connect_githubLink your GitHub from the client. Returns the FindAgent GitHub App install URL — install it and pick which repos to grant (private repos work; the pull is server-side).
• findagent_import_repoPull one of YOUR GitHub repos (via your connected token) and return the grounding — basics, languages, the tools it exposes, the hosts they reach, and the detected credential slots — so your own AI writes the listing. Read-only: nothing is persisted or executed.
• findagent_list_categoriesList the active marketplace categories so your AI can pick the right one for a listing.
• findagent_create_draftPersist the listing your AI wrote as a draft you own (free or paid). You review, price, and submit it in the wizard. Credential slots are declared (shape only) — never values.

These are a representative few — your client lists the full set of findagent_* tools once you connect the platform MCP.

You can still submit entirely in the browser — the submit wizard prefills what it can detect from your repo and you fill the rest. The MCP path just hands the writing to your own AI.

Agents are declarative: a manifest describes what an agent does and the shared runtime serves it — creators never ship executable config we run blindly. Learn the manifest shape, credential slots, and guardrails.

Install agents from the terminal with the FindAgent CLI, or call the install + manifest endpoints directly. Headless login uses the OAuth device flow.