Skip to main content
Glama

VibeView agent skills

npm GitHub MCP registry

Skills and an MCP server that let a coding agent verify its work on a live cloud device: upload a build, start an iOS, Android, Apple TV, Android TV or Roku session, drive the app by UI-tree refs, read logs, take screenshots, and stop when only a human can act.

VibeView streams iOS simulators, Android emulators, Apple TV and Android TV simulators, and Roku devices to a browser and to agents. No Mac, no local simulator, no device on the desk. Docs: AI agent control.

What your agent can do

  • Check its own change in your React Native app. It starts vibeview dev against your local Metro, so every edit hot-reloads on the cloud device, then it looks at the screen and the logs to see whether the change works.

  • Build the app in the cloud and run that build. React Native, Expo, Flutter, Capacitor, Kotlin Multiplatform or native iOS and Android: with a shell, it runs vibeview build --cloud (no Mac needed for iOS), then starts a device on the finished build.

  • Drive the app like a person. It reads the screen as a UI tree with short refs (@e7), taps, types and swipes by ref, presses TV remote buttons, opens deep links, and takes screenshots.

  • Write and run tests. It saves plain-language steps as a test, then runs one test or a whole group and reads which step failed and why. AI steps are in beta.

  • Hand off to you when a step needs a person, such as a login or a payment, with a link to the live session.

It works on iPhone, iPad, Android phones and tablets, Apple TV, Android TV and Roku. You can watch every session live in the dashboard while the agent works.

Related MCP server: pitbox

Install

You need the vibeview CLI, logged in:

npm install -g vibeview
vibeview login

Then pick whichever fits your agent.

Any agent that reads skills (Claude Code, Cursor, Codex, and others via the skills CLI):

npx skills add vibeview/skills

Claude Code, as a plugin (installs the skill and registers the MCP server in one step):

/plugin marketplace add vibeview/skills
/plugin install vibeview@vibeview

Claude Code, from the CLI you already installed (copies the skill into ~/.claude/skills/ and offers to register the MCP server):

vibeview agent-setup

Any MCP client: run vibeview mcp over stdio.

{
  "mcpServers": {
    "vibeview": { "command": "vibeview", "args": ["mcp"] }
  }
}

With Claude Code specifically: claude mcp add vibeview -- vibeview mcp.

What's here

Path

What it is

skills/vibeview-agent/SKILL.md

The agent workflow: build requirements, the dev loop, verifying with ui-tree, screenshots and logs, TV d-pad navigation, the human hand-off rule, cleanup, and the full command reference for both the CLI and MCP.

.claude-plugin/

Claude Code plugin manifest: the skill plus the vibeview mcp server.

.mcp.json

The same MCP server in the standard root-level form that plugin and MCP directories scan for.

Where this comes from

The canonical copy of the skill ships inside the vibeview npm package (vibeview agent-setup copies it from there). This repository mirrors it so agents and registries can discover it on GitHub. Versions track the CLI.

Issues and questions: support@vibeview.io.

Available Tools

38 tools
alertA

Inspect or respond to a visible system alert on iOS or Apple TV (get, accept, or dismiss). get returns the message and the button labels.

Inputs: action (required), button, session_id.

clear_textB

Clear the text in the currently focused input field.

Inputs: session_id.

create_testA

Create a test for an app from plain-language steps, one per line; a check is a step starting with "Verify". Returns the test id for run_test. The first run is driven by AI; once it passes, later runs replay the saved path.

Inputs: app (required), name (required), steps (required), group, +1 more.

dev_reloadA

Roku only: re-package the channel with the build command from vibeview.json (platforms.roku.build), upload it, and reload it into the session dev_start is holding (or, with no held session, this project's vibeview dev --detach session) — the MCP equivalent of pressing r in vibeview dev. Returns once the device restarts the channel (a few seconds); call ui_tree afterwards. Other platforms hot-reload through Metro and need no call.

Inputs: session_id.

dev_startA

Start a VibeView session held by this MCP server, and use it as the default session for every other tool call until dev_stop. By default it is a React Native dev loop: a debug build that loads its JS from your local Metro. Pass standalone: true to run any build as it is (a release, CI or cloud build) with no Metro. In a Flutter, Capacitor or native (not React Native) project the session is standalone automatically. Returns once the session accepts commands (device up, app installed), with the device it got (model, OS, form factor, posture) — call ui_tree next.

Inputs: platform (required), app, model, build_id, +3 more.

dev_stopA

Stop the dev-loop session started by dev_start.

Inputs: session_id.

dragB

Drag between two points, each an element ref (@e5) or raw coordinates (100,200).

Inputs: to (required), from (required), velocity, session_id, +2 more.

findC
Read-only

Find an element by text with optional spatial constraints (below/above/near).

Inputs: text (required), near, above, below, +1 more.

focusA
Read-only

TV: move focus to an element without activating it.

Inputs: ref (required), session_id.

get_runA
Read-only

A run's status and, when a test failed, which step and why. Pass the run id run_test or run_group returned.

Inputs: run_id (required).

list_appsA
Read-only

List your organization's uploaded apps: app id, name, platform and bundle id / package name. Sample apps (is_sample) are VibeView's examples, never this project's app. Pass an app id to dev_start (app) or list_builds.

Inputs: platform.

list_buildsA
Read-only

List an app's uploaded builds, newest first: id, version, build number, platform, kind (debug/release/unknown), upload time, note and retention (keep, expires_on as YYYY-MM-DD, expired; absent on older servers). Pass a debug build's id to dev_start as build_id.

Inputs: app (required).

list_device_modelsA
Read-only

List the device models you can start right now (model, OS version, platform, category, form_factor "foldable" for any device with a hinge — the iPhone Duo, Android foldables such as the Pixel 9 Pro Fold — and how many of each are free and busy). Pass a model name to dev_start to run on it; a model with none free queues until one is.

Inputs: device_type.

logsA
Read-only

Read recent app logs from the device — JS console output, native errors, and crash messages from the app under test (not full system logs). Check after a crash, a frozen screen, or an unexpected blank tree, and to confirm an expected log line fired. Pass the previous response's cursor as since to get only lines that arrived after it.

Inputs: tail, since, session_id.

long_pressB

Long-press (touch and hold) an element by its ref.

Inputs: ref (required), session_id, duration_ms.

open_urlA

Open a deep link in the app under test. The link goes to that app only, never a browser: on Android a link the app has no screen for fails; on iOS only the app's own URL scheme (e.g. myapp://settings) is accepted, and a web or system link is refused.

Inputs: url (required), session_id.

pressB

Press a device button or perform a system gesture (home, back, d-pad, etc).

Inputs: button (required), session_id.

relaunch_appA

Close the app under test and start it again: a cold start that keeps the app's data, e.g. to check that something persists. Its log keeps streaming across the restart. Not available on embed sessions.

Inputs: session_id.

rotateA

Rotate the device. A phone (and an Android foldable) toggles between portrait and landscape; the iPhone Duo turns a quarter clockwise, or by degrees 180/270. Reports the new orientation. Not available on TV devices.

Inputs: degrees, session_id.

run_groupA

Run every test of a group in order. With session_id it runs on that session; without it, on a new device of the app platform that VibeView starts and stops (it does not use a session you hold from dev_start; pass session_id to run on a session you hold). Returns a run id for get_run.

Inputs: group (required), ai_use, session_id, device_model, +1 more.

run_testA

Run a saved test. With session_id it runs on that session (the device you are driving). Without it, VibeView always starts a new device of the test app's platform, runs the test and stops the device (it does not use a session you hold from dev_start; pass session_id to run on a session you hold). Returns a run id for get_run.

Inputs: test (required), ai_use, session_id, device_model, +1 more.

screenshotA
Read-only

Capture a screenshot of the current screen. Use it to check how something LOOKS — the UI tree carries no colour, contrast, or stacking information, so a screen can pass every check the tree can express and still be visibly broken. Saved to disk; over MCP, pass inline: true to receive the image itself. On a phone or tablet its pixels are the coordinates tap/drag take, so a point read off it can be tapped as is.

Inputs: out, inline, session_id.

scrollA

Scroll the current view up, down, left, or right using a gesture preset.

Inputs: name (required), session_id.

scroll_toB

Scroll until an element matching the given text becomes visible.

Inputs: text (required), direction, session_id, max_scrolls, +1 more.

send_feedbackA

Send the VibeView team a bug report, a friction note or a feature idea about VibeView itself. Only for problems with VibeView itself (the CLI, MCP tools, devices, sessions, runs, dashboard, docs) — never for bugs in the user's own app. Show the user what you will send and get their OK first. Never include passwords, tokens, API keys, test data values or personal data. One report per problem; do not resend it. kind: bug (something broke), friction (it worked but was confusing, slow or awkward), idea (a missing capability). user_quote: the user's own words, verbatim — never a paraphrase. The report includes the session's device and status when it names one.

Inputs: kind (required), summary (required), details, session_id, +1 more.

set_languageA

Restart the app under test in another language — a language tag such as de, fr, ja or pt-BR, or 'default' for the device's language — e.g. to check translations or take localized screenshots. Only the app changes; the device, its status bar and other apps keep their language. iPhone, iPad, Apple TV and Android 13+; not Roku or embed sessions.

Inputs: language (required), session_id.

set_locationA

Set the device's simulated GPS location, for location-based flows. Phones, tablets and foldables (iOS and Android); not available on TV devices.

Inputs: latitude (required), longitude (required), session_id.

set_postureA

Fold or unfold a foldable device (form factor foldable in list_device_models — the iPhone Duo, Android foldables such as the Pixel 9 Pro Fold): a preset (closed, partial, open) or an exact hinge angle in degrees (angle 0-180) — pass exactly one. Reports the posture it settled in, the angle, and which screen is lit. Errors on a device without a hinge.

Inputs: angle, posture, session_id.

set_status_barA

Show a clean status bar for screenshots (9:41, full battery, full Wi-Fi and cellular signal) or put the device's own status bar back. It stays until you change it or the session ends. iOS phones and tablets only.

Inputs: preset (required), session_id.

shakeB

Shake the device once (phones, tablets and foldables)

Inputs: session_id.

stop_sessionA

Stop any session by id — e.g. one created through the API, the sandbox, or a collaboration link. For the session dev_start started, dev_stop does the same without an id.

Inputs: session_id (required).

swipeB

Perform a scrollbar-semantic swipe gesture in one of four directions.

Inputs: direction (required), distance, session_id.

tapB

Tap an element by ref (e.g. @e5), or raw coordinates (e.g. 100,200).

Inputs: target (required), session_id.

tap_focusedA

TV: move focus to an element and press SELECT in one step.

Inputs: ref (required), session_id.

typeA

Type text into the currently focused input field.

Inputs: text (required), session_id.

ui_treeB
Read-only

Fetch the ref-annotated UI tree for the current screen.

Inputs: session_id.

upload_appA

Upload an app build (zipped simulator .app for iOS/tvOS, .apk for Android/Android TV, or a Roku channel .zip) and return its app_id for dev_start. The MCP equivalent of vibeview upload-app.

Inputs: file (required), name.

waitB
Read-only

Wait for a specified number of seconds before continuing.

Inputs: seconds, session_id.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 38 tools

    Disambiguation4/5

    Most tools target a clearly distinct action+object: tap vs long_press vs drag vs type vs clear_text are unambiguous, and the descriptions actively separate the easily-confused trio swipe/scroll/scroll_to and find vs ui_tree. The remaining soft spots are dev_stop vs stop_session (overlap is explained in the text but still requires reading) and scroll vs swipe, which an agent could mix up without the descriptions. tap_focused vs focus vs tap on TV is handled well by explicit 'activates'/'without activating' wording.

    Naming Consistency4/5

    Everything is lowercase snake_case with a consistent terse style, and the namespaced families (dev_*, list_*, *_test, run_*/get_run, set_*) are predictable. Deviations are minor: some names are bare verbs (tap, press, wait, find, focus, rotate, shake, alert) while others are verb_noun (open_url, clear_text, scroll_to, relaunch_app), and a few are noun-first (ui_tree, logs, screenshot). Still readable and guessable.

    Tool Count3/5

    38 tools is a heavy surface, but the breadth is real: device interaction, device/system controls, session lifecycle, app/build catalogs, and the test runner are genuinely separate workflows. Individual tools each map to a distinct capability rather than padding, yet the set would benefit from tighter grouping or fewer near-duplicate scroll variants. Borderline over-scoped for a single MCP server.

    Completeness3/5

    Device control and inspection are well covered (input, gestures, rotation, posture, locale, location, logs, screenshot, tree, session start/stop/reload, app and build listing). The obvious gap is on the test surface: create_test/run_test/run_group/get_run exist but there is no list_tests, list_groups, update/delete_test, or run history listing, so an agent cannot discover existing tests or clean up. No session listing either — stop_session wants an id the server never helps you find.

    Maintenance

    ActivityActive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      Drives Android, iOS, smart TV (Android TV, Tizen, webOS), web and Bluetooth (BLE) devices from an agent: start a session, tap, type, swipe and assert by XPath, visible text, OCR or image, then save the steps as a replayable CSV test suite.
      530 PyPI
      99
      Apache 2.0
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables coding agents to run, drive, and debug apps across Android emulators, Apple simulators, devices, and browsers from one tool, with token-lean UI snapshots, scripted flows, checks, and local service orchestration.
      9,401 npm
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables an MCP-connected agent to start and control real Android emulator and iOS Simulator sessions on GitHub Actions or locally, observe screens and UI trees, perform touch and text actions, and capture screenshots, recordings, logs, crashes, and evidence.
      Apache 2.0