asomaster-mcp
# ASOMaster MCP server
`asomaster-mcp` connects AI agents (Claude Code, Codex, Cursor, Claude Desktop, Windsurf, OpenCode or any
stdio MCP client) to [ASOMaster](https://asomaster.app), a desktop ASO app for the App Store and Google Play.
Your agent can read keyword boards, check ranks, find and localize keywords, fetch real Apple Ads or
Google Ads popularity, watch rivals and run the listing studio, with every change showing up live in the
app. The bridge talks to the app over a private local socket (a named pipe on Windows); nothing goes
through our servers.
Full docs: <https://asomaster.app/docs/mcp/> (Markdown for agents: <https://asomaster.app/docs/mcp.md>)
## Install
`asomaster-mcp` ships inside the ASOMaster app (this repository is its source). [Download ASOMaster](https://asomaster.app) for macOS
(Windows beta), activate a license, then either:
- **From the app:** Settings › MCP clients › Install, for Claude Code, Codex, OpenCode, Cursor, Claude
Desktop or Windsurf.
- **By hand:** point your client at the bridge.
| System | Bridge path |
|---------|-------------|
| macOS | `/Applications/ASOMaster.app/Contents/MacOS/asomaster-mcp` |
| Windows | `%LOCALAPPDATA%\ASOMaster\asomaster-mcp.exe` |
Claude Code, as a plugin (folder [plugin/](plugin/)):
```
/plugin install asomaster --marketplace tarek369/asomaster-mcp
```
Or add the server directly:
```sh
claude mcp add --scope user asomaster -- /Applications/ASOMaster.app/Contents/MacOS/asomaster-mcp
```
Cursor, Claude Desktop, Windsurf and other JSON configs ([examples/mcp.json](examples/mcp.json)):
```json
{
"mcpServers": {
"asomaster": {
"command": "/Applications/ASOMaster.app/Contents/MacOS/asomaster-mcp",
"args": []
}
}
}
```
Codex (`~/.codex/config.toml`):
```toml
[mcp_servers.asomaster]
command = "/Applications/ASOMaster.app/Contents/MacOS/asomaster-mcp"
args = []
```
The app doesn't need to be open; the bridge starts it. Supported MCP protocol versions: 2024-11-05,
2025-03-26, 2025-06-18 and 2025-11-25.
## What agents can do
- **Read:** apps, keyword boards per market, keyword detail (rank history, apps ranking), rank alerts,
audit score, ratings, rival listing changes (`get_overview`, `get_project_board`, `get_term_detail`,
`list_rank_alerts`, `list_rival_changes`, `get_audit`, `get_ratings`).
- **Keywords:** suggest, add, localize, edit and remove keywords; popular App Store searches around a
seed word (`add_keywords`, `edit_keyword`, `popular_searches`).
- **Demand:** real popularity from Apple Ads or Google Ads Keyword Planner, kept separate from
ASOMaster's own estimates (`connect_popularity`, `get_popularity_status`).
- **Audit:** run the store listing audit per country, read the score, issues and competitors, and
dismiss competitors that don't fit (`run_audit`, `get_audit`, `dismiss_audit_competitor`).
- **Listings:** draft, compare, choose and localize titles, subtitles, keyword fields and descriptions;
run a final review after translating, undo edits, keep a don't-target list and preview changes
(`get_optimizer`, `optimizer_review`, `get_publish_preview`).
- **Pricing:** read App Store and Google Play prices per country and plan new ones, with local price
suggestions (`get_pricing`, `plan_pricing`).
- **Screenshots:** edit slides, import store screenshots, write and translate headings, render and
export a fastlane layout (`get_screenshots`, `screenshots_render`, `screenshots_export`).
- **Jobs:** store searches and AI runs are queued; `wait_for_jobs` waits until they finish.
## Safety
- Anything that changes a live store page (publishing a listing, prices, screenshots or a replacement
app name) needs your go-ahead twice: the agent must pass your explicit approval, and the app shows a
native confirmation only you can answer. Nothing is submitted for review on the App Store; on Google Play
changes wait for you to send them, except for apps whose changes Play always sends for review.
- Credentials are passed as file paths, stored in the OS keychain and never returned by a tool.
- Local only: owner-only Unix socket on macOS, per-user named pipe with a token on Windows.
## Source, Linux build and Docker
This repository holds the bridge's source (Rust, MIT). `cargo build --release` builds it; `cargo test`
runs its tests. The [Dockerfile](Dockerfile) builds and runs it so MCP directories can check that the
server starts and lists its tools. ASOMaster itself runs on macOS and Windows; without the app the
bridge still answers `initialize` and `tools/list`, and every tool call returns `app_unavailable`.
## Requirements
ASOMaster with an active license (subscription: 15.99 EUR per month or 49 EUR per year). Without one,
tools answer `license_required`.
## Support
Questions and bug reports: open an issue here or see <https://asomaster.app/support/>.
ASOMaster is made by Cressu OÜ.
TDQS
Scored across 94 tools
Most tools have a narrow, well-described job and the descriptions cross-reference each other (e.g. 'then wait_for_jobs and get_reviews'), which helps. But the analytics reads overlap heavily — get_dashboard, get_app_insights ('same measures as get_dashboard'), get_project_board, get_audit and get_change_impact all surface rank/visibility data for overlapping scopes, and choosing the right one for a question requires careful reading rather than name alone.
Dominant pattern is verb_noun snake_case (get_/list_/add_/remove_/set_/run_/apply_) with consistent domain prefixes for the large subsystems (optimizer_*, screenshots_*, *_jobs, *_console, *_popularity). Minor deviations like 'popular_searches' (noun-only) and 'dont_target' (verb-only) are readable but break the pattern.
94 tools is an extreme surface for one server, far beyond the 3-15 sensible range. The domain is genuinely broad (keywords, rankings, listings, screenshots, pricing, reviews, console auth, AI, jobs), yet the volume means an agent must hold an enormous mental map just to pick a tool, and many entries are single-field setters that could be folded into fewer parameterized tools.
Lifecycle coverage is unusually thorough: project/keyword/folder CRUD, rank checks, optimizer drafts through live publish, screenshot design through live upload, pricing plans through live publish, reviews, console and popularity connections, and job control. Minor gaps exist (e.g. no tool to clear or delete a saved review reply draft, no generic 'get live listing' read separate from get_publish_preview/resync_project), but agents can route around them.