rollout MCP server
Commit Nov 11, 2026Lets an AI agent manage Android releases, track staged rollouts, monitor vitals, and manage Play Store listings via the Google Play Developer APIs.
Also known as Limetric/rollout
- Stars
- 1
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 11, 2026
- License
- Apache-2.0
What is rollout MCP?
rollout lets an AI agent manage Android releases, track staged rollouts, monitor vitals, and manage Play Store listings via the Google Play Developer APIs.
Summary generated from the project's README
About rollout MCP server
Managing Android releases and monitoring app health from within an AI agent is what rollout enables. Developers and DevOps engineers can use this tool to push builds to specific tracks, monitor staged rollout percentages, and investigate performance issues like crash rate spikes without navigating the Google Play Console. It connects directly to the Android Publisher and Play Developer Reporting APIs from your local machine, ensuring your data stays private without a relay server.
The server provides 53 distinct tools covering tracks, releases, store listings, in-app products, subscriptions, reviews, and Android vitals. It operates via stdio for MCP hosts like Claude Desktop or Cursor, and can be configured using a service account JSON key via environment variables. To prevent accidental changes, every mutating action such as halting a rollout or editing a listing requires a preview and a confirmation token before the change is applied.
What you can do with it
- 1Monitor staged rolloutsAsk your AI agent to check the current progress and percentage of a production rollout.
- 2Investigate app stabilityRequest an analysis of why crash rates have increased for a specific version code.
- 3Manage store presenceHave an agent update your app's store listing or manage user reviews.
- 4Safely execute releasesUse the confirmation system to ensure no release changes happen without explicit approval.
Written from the project's README. Every use case is checked against a quote from it.
Configuration
PLAY_PACKAGE_NAMEPLAY_SERVICE_ACCOUNT_FILEROLLOUT_TOKEN_STORE
Automated checks
Ran Nov 15, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablelimetric/rollout is public and reachablegithub.com/limetric/rollout ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)Apache-2.0spdx.org/licenses/Apache-2.0.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-11github.com/limetric/rollout ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubrollout
Ship an Android release without opening Play Console. rollout is the CLI
and MCP server that lets you (or your favorite AI agent) push a build to a
track, watch a staged rollout climb, halt it if things look wrong, and answer
"why did our crash rate spike" — all from a terminal instead of a maze of
dashboard tabs. Store listings and user reviews come along for the ride. It
talks straight to the Android Publisher and Play Developer Reporting APIs from
your own machine — no relay server sitting between you and your app's data.
It ships as a single binary with two front-ends over one shared set of tools:
- CLI —
rollout play releases,rollout play tracks,rollout play promote …. Scriptable, pipeable intojq, usable in CI. This is what the bundled agent skill drives. - MCP server —
rollout mcpserves the same tools over stdio to MCP hosts (Claude Desktop, Cursor, …), so an agent can manage your release the same careful way you would.
rollout play releases --format table
track | status | version_codes | rollout_percent | name
------------+------------+---------------+-----------------+------
production | completed | ["41"] | | 1.4.0
production | inProgress | ["42"] | 10 | 1.5.0
beta | completed | ["43"] | | 1.6.0
Nothing mutates on the first ask. Every write — halting a rollout, promoting a build, editing a listing — comes back as a preview and a confirm token first. You (or the agent) apply it on purpose, or it expires.
Platforms
| Platform | Setup guide | Surface |
|---|---|---|
| Google Play | docs/play.md |
53 tools — tracks, releases, listings, in-app products and subscriptions, reviews, Android vitals, and the CSV report exports |
The platform registry exists so a second store (App Store Connect is the obvious one) is a new file rather than an edit to the shared auth, config, and safety plumbing.
Install
Three ways in, no wrong answer:
-
Homebrew (macOS/Linux):
brew install Limetric/tap/rollout -
Prebuilt binary — grab one from the releases page.
-
Build from source:
go build -o build/rollout .
Quick start
rollout login play # guided setup + live check
rollout config play set-package com.example.app # so --package can be omitted
rollout play tracks
CI skips the wizard:
export PLAY_SERVICE_ACCOUNT_FILE=/path/to/key.json
export PLAY_PACKAGE_NAME=com.example.app
rollout doctor play
Don't forget to invite the credential in Play Console → Users & permissions
— authenticating isn't the same as having access. Details in
docs/play.md.
Integrations
As an MCP server
rollout mcp serves the same tools over stdio, under a platform prefix —
play_tracks, play_releases, play_reviews, … Point your host at the
binary and pass credentials through the environment, no config file needed:
{
"mcpServers": {
"rollout": {
"command": "rollout",
"args": ["mcp"],
"env": {
"PLAY_SERVICE_ACCOUNT_FILE": "/path/to/key.json",
"PLAY_PACKAGE_NAME": "com.example.app"
}
}
}
}
A platform whose credentials don't resolve is skipped with a note on stderr rather than taking the whole server down, so an unconfigured second store never blocks the first.
As a Claude Code plugin
The repo bundles a skill (plugins/rollout/skills/rollout/SKILL.md) that
teaches an agent when and how to drive the CLI. If you installed rollout via
Homebrew and don't have the repo cloned, install it as a plugin instead:
/plugin marketplace add Limetric/rollout
/plugin install rollout@rollout
As a Codex plugin
Codex reads the same skill through its own plugin manifest. Install the
rollout binary using one of the install options, make sure it is
on your PATH, then add the plugin:
codex plugin marketplace add Limetric/rollout
codex plugin add rollout@rollout
Complete the quick start to configure Google Play access, then start a new Codex thread to use the skill. For example: "What version is live on production?"
With the skills CLI
For any other agent the skills CLI
supports — Cursor, Gemini CLI, GitHub Copilot, OpenCode, … — install the same
skill straight from the repo. As with the plugins, the rollout binary has to
be installed and on your PATH:
npx skills add Limetric/rollout --skill rollout # into the current directory
npx skills add Limetric/rollout --skill rollout --global # for all directories
skills asks which agents to enable it for; pass --agent <name> to pick one
up front. Run npx skills update later to pull in the latest skill.
Concepts
Writes preview first
Every mutating call returns a preview and a confirm token before it touches anything:
rollout play release halt --track production
# → preview + confirm token, valid for 10 minutes
rollout confirm <token> # applies it (or re-run with --confirm <token>)
rollout audit # log of every write rollout has applied
Writes are transactional — a release write opens a fresh edit, mutates one release, validates, and commits, deleting the edit on any failure, so nothing is left half-staged. It also reads the track first and touches only the release it names, so an in-progress rollout is never dropped by a write that didn't mention it. Halting a rollout, completing a production rollout, and deleting a listing or image type ask for a second confirmation, since none of those undo with the opposite command.
Guard rails are optional and off by default — a production lock, a rollout
percentage ceiling, a blocked-operations list — and they're re-checked at
confirm time, not just when the preview is issued. See
docs/play.md for the full list.
Where the refresh token lives
rollout login play writes the refresh token to a per-platform token
store (0600, under the config directory), never to config.toml and
never to an environment variable:
rollout doctor play # store location, writability, sign-in age
export ROLLOUT_TOKEN_STORE=/path/to/tokens # containers/CI: mount a writable volume
A service-account JSON key skips the token store entirely and wins when both credential modes are configured — it's what CI and Google's own docs recommend for the Publisher API.
Defaults and output
rollout config play set-package com.example.app # drop --package everywhere else
rollout play releases --format table # or --format csv for a spreadsheet
rollout play vitals --metric crashrate --dimension versionCode | jq '.rows'
Reads print JSON by default, so they pipe straight into jq. Human-facing
output — login, doctor, config show, table headers, write previews — is
colored on a terminal and plain everywhere else, so scripts and rollout audit stay byte-identical either way. Override with --color always|never|auto;
NO_COLOR and TERM=dumb are honored too.
Documentation
docs/play.md— setup, tool coverage, release semantics, troubleshootingdocs/reporting.md— vitals metric sets, dimensions, thresholdsdocs/reports.md— the CSV report exports: installs, ratings, financialsdocs/name-map.md— CLI ↔ MCP name map
See AGENTS.md for the contributor workflow and conventions.
License
Apache-2.0. See LICENSE.
rollout MCP questions
How does the server handle destructive actions like halting a rollout?
Every mutating call returns a preview and a confirm token first; you must apply it on purpose using the confirm command.
What kind of data can I access from Google Play?
You can access tracks, releases, listings, in-app products, subscriptions, reviews, Android vitals, and CSV report exports.
Is it safe to let an AI agent make changes?
Yes, writes are transactional and include guard rails like production locks and rollout percentage ceilings that are re-checked at confirmation.