Skip to content
FindAgent
Sign in
Documentation menu
Using agents

Install an agent

The guided web installer is the quickest way to add any agent to your LLM client — it writes the exact files the agent needs, with nothing to set up. A command-line tool covers the same flow for scripted installs.

Install from the web (recommended)

Every agent has a guided installer that walks you through adding it to your client. It produces the same files described on this page — no tooling required, and it works today.

  1. Open the agent. From an agent's page, choose Install to open its guided installer.
  2. Pick your client. Select Claude, ChatGPT, Gemini, Cursor, or a custom runtime — the installer shows the exact files for that target.
  3. Copy in the files. Follow the on-screen steps to drop the files into your client. That's it — the agent is ready to use.

Browse the registry and open any agent to start, or jump straight to the marketplace.

Also available: the CLI
Prefer the terminal? The findagent CLI installs the same files from the command line and is handy for scripted or repeatable setups — run it on demand with npx @findagent/cli, no global install needed. The web installer produces the exact same result.

Sign in (device flow)

Installing an agent ties access to your FindAgent account, so the first step is signing this machine in. The CLI uses a browser device flow — no token to copy by hand:

terminal
npx @findagent/cli login

The CLI prints a short code and opens a confirmation page. Approve it in the browser and the machine is authenticated. You can check who is signed in with findagent whoami and sign out with findagent logout.

Headless / CI
On a machine with no browser, paste a token directly: findagent login --token <jwt>.

Install an agent

Install by slug, optionally pinning a version, and pick the client you want it written for:

terminal
npx @findagent/cli install atlas --llm claude

Pin a specific version with the slug@version form:

terminal
npx @findagent/cli install atlas@1.0.0 --llm cursor

Useful flags:

  • --llm <target> — which client to write files for: claude, chatgpt, gemini, cursor, or custom.
  • --dir <path> — the project directory to install into (defaults to the current directory).
  • --dry-run — print exactly what would be written without touching the filesystem.

Per-client adapters

The CLI knows how each client expects an agent to be wired, and writes the right files for the target you choose:

  • Claude — Claude Desktop and Claude Code.
  • ChatGPT — for clients that support custom instructions / MCP.
  • Gemini — Gemini CLI and compatible clients.
  • Cursor — the in-editor agent config.

For a doer agent, this includes the manifest and any tool wiring so the local runtime can execute the agent's declared tools. Recipe agents are written as the system prompt and example flows.

Credentials

If an agent calls an external service it declares credential slots in its manifest. Store a value for a slot locally — it never leaves your machine and is only attached to requests whose host the slot allows:

terminal
npx @findagent/cli secrets set <ref> <value>
npx @findagent/cli secrets list
npx @findagent/cli secrets rm <ref>

See the manifest spec for how credential slots bind a secret to an allowed host.

Manage what you have

List the agents you have purchased or installed:

terminal
npx @findagent/cli list

The full command and flag reference lives on the CLI reference page.