Connect Gemini¶

Gemini's equivalent of a Skill is a Gem (custom instructions). To actually run searches you need a way to make HTTP/MCP calls — that's Gemini CLI or the Gemini API. The consumer Gemini app can't call custom MCP servers or arbitrary HTTP, so a Gem there can only describe the behavior, not execute it.

Option A — Gem¶

Install. A Gem is created in the Gemini app UI, not from a file. Create a new Gem and paste the instructions from the connectors repo (skills/gemini/):

Name: agentfinder
Instructions: copy the body of that file — it tells Gemini to show a menu of named Agent Finders, remember the choice, present the ranked results, and never auto-install.

Pair the Gem with the Gemini CLI MCP setup below (or the Gemini API) so it can make the call.

How to invoke it¶

With the Gem selected (and the CLI/API able to call Agent Finder), ask in plain language — "Find me a tool that summarizes long PDFs." Gemini shows a menu of Agent Finders the first time (and remembers your pick), queries it, and lists the matches.

Option B — Remote MCP connector (Gemini CLI)¶

Add the server to your Gemini CLI settings.json:

{
  "mcpServers": {
    "agent-finder": {
      "httpUrl": "https://agentfinder.github.com/api/v1/mcp"
    }
  }
}

Use httpUrl for a Streamable-HTTP server, url for SSE, or command/args for a local/stdio server. Restart the CLI; the Agent Finder search tool is then available. (The Gemini API can connect to remote MCP servers directly, or you can expose POST /search as a function.)

How to invoke it¶

In the Gemini CLI, ask it to find a capability for your task; it calls the search tool and lists matches. Pair with the Gem (Option A) so it offers the finder menu, remembers your choice, and never auto-installs.

Endpoint¶

Examples use GitHub's Agent Finder (https://agentfinder.github.com/api/v1); Hugging Face Discover (https://huggingface-hf-discover.hf.space/search) works the same way. Point at either — or any compliant ARD discovery service — see Endpoints.