About TapTry
TapTry is a CLI and MCP server for uploading Android builds and sharing browser-accessible links. It also lets MCP clients operate an uploaded app on a cloud Android device, including screenshots, UI interaction, and log retrieval.
Key features
- 01Upload APK, AAB, and related packages
- 02Malware scanning before links activate
- 03Public, unlisted, or private visibility
- 04GitHub pull request preview workflow
- 05Cloud device interaction and screenshots
- 06Read UI elements and app logs
Use cases
- 01Share an Android build without installation
- 02Publish preview links for pull requests
- 03Test an uploaded app on a cloud device
- 04Inspect crashes and recent app logs
MCP support
The MCP server exposes tools to upload, inspect, list, and delete apps, as well as start device sessions, interact with apps, and retrieve screenshots, UI elements, and logs. It connects to the TapTry API using an API key.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
npx -y taptryTAPTRY_API_KEYSecretA TapTry API key - create one at https://taptry.in/settings/api-keysLocal servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
14Saved tool list · updated 1w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| ref | string | No | A stable id for this app, e.g. the repo name, so re-uploads update it instead of creating a new app. |
| name | string | No | App name. Default: from the file name. |
| file_path | string | Yes | Absolute path to the app file, e.g. app/build/outputs/apk/release/app-release.apk. |
| visibility | enum | No | Default public. Unlisted suits a build to share with specific people.public | unlisted | private |
| wait_seconds | number | No | How long to wait for the checks before returning (0-600, default 90). Use get_status after that. |
Details
From the project READMEtaptry
Turn an Android build into a link anyone can run in their browser - no install, no sideloading. From your terminal, from CI on every pull request, or from your coding assistant.
npx taptry upload app/build/outputs/apk/release/app-release.apk
# Uploading app-release.apk (18.2 MB)…
# Created My Game - https://taptry.in/app/my-game
# Waiting for the malware scan and checks (this can take a few minutes)…
# Checking for malware…
# Ready
# https://taptry.in/app/my-game
Takes .apk, .aab, .apks, .xapk and .apkm, up to 500MB. Every upload
is malware-scanned before its link works. Needs Node 20+; no dependencies.
Setup
- In TapTry, open Settings > API keys and create a key.
- Either save it on your machine -
npx taptry login- or setTAPTRY_API_KEY(what CI should do).
CLI
taptry upload <file> [--name <name>] [--ref <ref>] [--unlisted | --private] [--no-wait] [--timeout <s>] [--json]
taptry status <app> | --ref <ref>
taptry list [--ref <ref>]
taptry delete <app> | --ref <ref> [--yes]
taptry login | logout | whoami
taptry mcp
--refis your own id for an app, e.g.github:me/game#42. Uploading again with the same ref replaces that app's file and keeps its link - use it for anything you upload repeatedly.- Visibility: public by default (on your TapTry profile, anyone can
run it).
--unlisted: listed nowhere, and the link printed is a secret link only the people you send it to can open.--private: only you. uploadwaits for the checks by default and prints the link on stdout, soURL=$(taptry upload app.apk)works.--jsonprints the whole result.deleteasks first; without a terminal it needs--yes.
Exit codes: 0 ok · 1 error (network, auth, refused request) · 2 bad
usage · 3 the app failed its checks (malware, not an app, …) · 4 timed out
waiting for the checks.
GitHub Action: a preview link on every pull request
Each pull request gets one TapTry app and one comment with its link, updated on every push and deleted when the PR closes. Previews are unlisted by default: anyone with the link in the PR can run them, and they never show up on your TapTry profile.
# .github/workflows/taptry-preview.yml
name: TapTry preview
on:
pull_request:
types: [opened, synchronize, reopened, closed]
permissions:
contents: read
pull-requests: write
jobs:
preview:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
if: github.event.action != 'closed'
- uses: actions/setup-java@v4
if: github.event.action != 'closed'
with:
distribution: temurin
java-version: 17
- name: Build
if: github.event.action != 'closed'
run: ./gradlew assembleDebug
- uses: Romildoescoding/taptry-cli@v1
with:
api-key: ${{ secrets.TAPTRY_API_KEY }}
file: app/build/outputs/apk/debug/app-debug.apk
Add your key as the repository secret TAPTRY_API_KEY. Pull requests from
forks don't get secrets, so their previews are skipped with a warning.
| Input | Default | |
|---|---|---|
api-key | - | Required. |
file | - | The build output. A glob is fine if it matches exactly one file. |
name | from the file | App name, set on the first upload. |
ref | github:<repo>#<PR> or github:<repo>@<branch> | Same ref, same app. |
visibility | unlisted | unlisted (anyone with the link, listed nowhere), public or private. |
wait / timeout | true / 900 | Wait for the checks. |
comment | true | Keep one PR comment up to date. |
delete-on-close | true | Delete the preview when the PR closes. |
fail-on-error | true | Fail the step when the build is refused. |
Outputs: url, app-id, slug, status, ready.
Outside pull requests (e.g. on: push to main), it keeps one app per
branch: github:<repo>@main.
MCP server: "build it and give me a link"
taptry mcp is a Model Context Protocol
server for Claude Code, Cursor, Claude Desktop and other MCP clients. Tools:
upload_app, get_status, get_link, list_apps, delete_app. The
assistant is told to ask before uploading or deleting, and delete_app
refuses unless confirm: true is passed.
Claude Code
claude mcp add taptry -e TAPTRY_API_KEY=tt_your_key -- npx -y taptry mcp
Cursor (~/.cursor/mcp.json) or Claude Desktop
(claude_desktop_config.json):
{
"mcpServers": {
"taptry": {
"command": "npx",
"args": ["-y", "taptry", "mcp"],
"env": { "TAPTRY_API_KEY": "tt_your_key" }
}
}
}
After taptry login you can leave out the env block.
Claude Code plugin - the MCP server and the TapTry skill in one install (set TAPTRY_API_KEY in your environment first):
/plugin marketplace add Romildoescoding/taptry-cli
/plugin install taptry@taptry
Let your assistant test the app on a real device
The MCP server can also run your uploaded app on a cloud Android device and
let the assistant use it: start_session, then screenshot (it sees the
screen) and get_ui (the elements, with coordinates), tap (by label or
position), type_text, press_key, swipe, get_logs (crashes and
recent log lines) and end_session. So "upload the build, open it, sign up
with a test account and tell me if anything crashes" works end to end. Each
session uses one of your runs and ends after 5 minutes without a call; at
most 2 run at once.
Data and privacy
Everything here talks to one service: the TapTry API at https://api.taptry.in, using your API key.
- Sent to TapTry: app files you upload (
.apk,.aab,.apks,.xapk,.apkm), the app name and settings you give, and - in device sessions - the taps, swipes, keys and text you send to your app. - Read from TapTry: your account name and plan (
whoami), your apps and their status, and in device sessions the screenshots, on-screen elements and logs of your own app. Screenshots and logs can contain whatever your app displays or logs. - Kept by TapTry: uploaded apps until you delete them (
taptry delete, or in the web app); run counts and analytics for your apps. Device-session screenshots, UI reads and logs are returned to you and not stored. Uploaded files are malware-scanned by VirusTotal. Privacy policy · Terms. - Stored locally:
taptry loginsaves your API key in your user config folder (mode 0600). Nothing else is written.
API
Everything here wraps TapTry's public HTTP API at
https://api.taptry.in/api/v1 (bearer-token auth with the same key). A
full reference for using it directly is coming to TapTry's docs.
License
MIT
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.