Ramus Android

Ramus android-use

Android-use for coding agents, powered by Ramus. Give Claude Code, Codex, Cursor, Gemini CLI or any agent with a shell a disposable cloud Android emulator it can see, tap, type on, screenshot and read logs from, then watch or take over in your browser.

This plugin connects to the Ramus hosted service. It needs a Ramus account (free during alpha) or a 60-minute CLI trial.

Install

Claude Code, as a plugin (the android-use skill plus the Ramus MCP server; you are asked for an optional API key):

/plugin marketplace add ramus-dev/android-use/plugin install ramus@ramus

Cursor, Codex, GitHub Copilot, Kiro and other Agent Plugins clients: install this repository as a plugin (plugin.json at the root).

Any agent that supports skills, via the skills CLI:

npx skills add ramus-dev/android-use

Claude Code, by hand: copy skills/android-use/SKILL.md to .claude/skills/android-use/SKILL.md.

Gemini CLI, as an extension (bundles the skill and the Ramus MCP server):

gemini extensions install https://github.com/ramus-dev/android-use

Any MCP client (Cline, Claude Desktop, Cursor, Windsurf, VS Code), as a remote server. Streamable HTTP, no local process:

{  "mcpServers": {    "ramus": {      "url": "https://api.ramus.dev/api/mcp",      "headers": { "Authorization": "Bearer <RAMUS_API_KEY>" }    }  }}

Create the key at https://ramus.dev/settings. The server exposes android_* tools (start a session from a demo app or pull request, snapshot the UI tree, tap, type, press, wait, screenshot, logs, share); list them with tools/list. Local APK upload and the dev loop need the CLI below.

The trial needs no account:

npx ramus-cli trialnpx ramus-cli session start --apk app-debug.apk --wait

ChatGPT and Codex (OpenAI plugin)

The same repository is the OpenAI plugin package: plugin.json carries the directory listing under extensions.com.openai, mcp.json points at the remote MCP server (people connect by signing in to Ramus; no API key), and the skills declare where they run in skills/*/agents/openai.yaml. The OpenAI package ships only android-mcp (MCP tools, ChatGPT and Codex): android-use drives the CLI through the shell, which OpenAI's skill scan flags, so the OpenAI packaging workflow leaves it out.

Maintainers can build the upload for the OpenAI plugin portal ("With MCP" path) using the manual Package OpenAI plugin workflow:

1. Open Actions → Package OpenAI plugin → Run workflow and choose the branch or tag to package.
2. After it succeeds, download the ramus-openai-upload artifact from the run and unpack it.
3. Upload the enclosed ramus-openai-plugin.zip to the OpenAI plugin portal.

The workflow checks the listing against OpenAI's limits and packages only the OpenAI manifest, MCP configuration, license, logo and android-mcp skill. It runs only in GitHub Actions when a maintainer requests an upload; the installed plugin does not run the packaging code.

How it works

The device is a real Android emulator running stock Google images on the Ramus fleet, streamed over WebRTC. The skill teaches the agent the observe, act, verify loop: read a snapshot of the UI tree, act on a referenced element, then check the result before moving on.

Links

• Docs: https://ramus.dev/docs/agents

• Agent reference: https://ramus.dev/agents.md

• MCP endpoint (Streamable HTTP, API key): https://api.ramus.dev/api/mcp

• CLI on npm: https://www.npmjs.com/package/ramus-cli

• Privacy policy: https://ramus.dev/privacy

• Terms: https://ramus.dev/terms

• Acceptable use: https://ramus.dev/acceptable-use

• Support: [email protected] · Security: [email protected]

Android only. Free during alpha. For testing apps you build or are authorized to test.

SKILL.md is generated from the Ramus product contracts; the canonical copy is served at https://ramus.dev/skill.md.