lumi-mcp
Privacy-preserving MCP app finder for Lumi Studio's live iOS apps, across all 50 Apple locales.
Documentation
Lumi App Finder MCP
Lumi App Finder gives AI assistants one read-only tool for matching a user's
task or buyer need to a verified live Lumi Studio iOS app.
It covers 46 apps × all 50 Apple locales. Every result includes
editorially localized context, the purchase model, a detailed guide, and a
direct App Store link for the matching storefront.
Hosts that support the stable MCP Apps extension render the matches as
interactive, localized cards with one-click App Store and guide actions.
Other hosts receive the same complete text and structured-data fallback.
> First-party disclosure: Lumi Studio develops every listed app. Results
> are transparent publisher-authored text matches, not measured search volume,
> independent rankings, reviews, or user endorsements.
Agent Skill
AI hosts that support the open Agent Skills specification can install the same
50-locale, first-party catalog as an offline, progressively loaded skill:
gh skill install alice51849/lumi-mcp lumi-app-finder --scope userGitHub CLI 2.90 or newer installs to GitHub Copilot by default. Add
`--agent claude-code`, `--agent cursor`, `--agent codex`, or
`--agent gemini-cli` for another supported host. The installed skill reads only
the requested locale's 46-record snapshot and requires no account, API key,
network request, or executable script at runtime.
The version-pinned skill also works with the Vercel Skills CLI and its supported
agents:
npx -y skills@1.5.19 add https://github.com/alice51849/lumi-mcp --skill lumi-app-finder -g -yThe third-party installer reports anonymous installation telemetry by default
under the skills.sh CLI policy. The installed Lumi
App Finder skill itself remains offline and contains no analytics.
The public skill is discoverable on its
which reads the same first-party `SKILL.md` from this repository.
Lumi App Finder is also listed in the independently maintained
which links to the published skill files.
Tool
`find_ios_apps`
Inputs:
- `query` — task, app name, or buyer need in any supported language.
- `locale` — one of Apple's 50 supported locale codes; defaults to `en-US`.
- `limit` — 1–10 matches; defaults to 5.
The release snapshot is generated from the authoritative live publisher
registry and rejected unless it contains exactly 46 unique App Store IDs × 50
Apple locales = 2,300 complete records. The runtime reads only that
digest-verified bundled snapshot: it makes no network request, and user query
text never leaves the local MCP process.
Install
Find `io.github.alice51849/lumi-app-finder` in clients or registries that use
the official MCP Registry, or use a version-pinned installer:
The VS Code and Cursor links run the zero-dependency server from the latest
verified public GitHub release through `npx`; Node.js 20 or newer is required.
The same configuration can be added manually:
{
"servers": {
"lumi-app-finder": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"https://github.com/alice51849/lumi-mcp/releases/latest/download/lumi-app-finder-npx.tgz"
]
}
}
}The MCPB uses Node's stdio transport and needs no account, API key, external UI
runtime, or manual configuration.
The source includes a multi-architecture OCI release path. Main does not
advertise an OCI tag as installable until release publication verifies an
existing public GHCR package, a candidate digest, ownership labels, SBOM,
provenance, and a 50-locale MCP smoke test, then promotes that same digest to
the public semver tag and Registry.
For Cline, use the exact configuration and verification steps in
50-locale coverage
`ar-SA`, `bn-BD`, `ca`, `cs`, `da`, `de-DE`, `el`, `en-AU`, `en-CA`,
`en-GB`, `en-US`, `es-ES`, `es-MX`, `fi`, `fr-CA`, `fr-FR`, `gu-IN`, `he`,
`hi`, `hr`, `hu`, `id`, `it`, `ja`, `kn-IN`, `ko`, `ml-IN`, `mr-IN`, `ms`,
`nl-NL`, `no`, `or-IN`, `pa-IN`, `pl`, `pt-BR`, `pt-PT`, `ro`, `ru`, `sk`,
`sl-SI`, `sv`, `ta-IN`, `te-IN`, `th`, `tr`, `uk`, `ur-PK`, `vi`,
`zh-Hans`, `zh-Hant`.
MCPB display metadata and the catalog output both use these localized
resources.
Development
npm ci
npm run catalog:gate
npm run build:ui
npm run build:skill
npm test
npm run smoke
npm run benchmark
npm run validate
npm run validate:registry
npm run pack:mcpbRefresh the bundled catalog and 50 locale resources from the public
`ios-app-guide` source:
npm run sync:catalogVersion tags publish the MCPB and attested multi-architecture OCI image, then
update the official Registry through GitHub OIDC without long-lived registry
credentials. Registry publication is the final step and cannot run before all
artifact gates succeed.
Privacy
See PRIVACY.md. The runtime has zero telemetry, no analytics
or tracking, no network requests, and no query, IP-address, or user-agent
logging.
Security
Report vulnerabilities privately through the
security policy, not through a public issue.
License
MIT. Third-party language-data notices are in
THIRD_PARTY_NOTICES.txt; bundled MCP App library
licenses are in MCP_APP_NOTICES.txt.
Frequently asked questions
What is lumi-mcp?
lumi-mcp is Privacy-preserving MCP app finder for Lumi Studio's live iOS apps, across all 50 Apple locales.
How do I install lumi-mcp?
Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.
Is lumi-mcp open source?
Yes — it is hosted on GitHub at https://github.com/alice51849/lumi-mcp.
Related MCP tools
Wondel.ai Agent Skills — Business, Marketing, UX & Coding Frameworks from Bestselling Books. 50 skills + 12 guided journeys for Claude Code, Codex, Cursor & other agentskills.io agents.
One place to manage & connect to all your MCP servers
提取抖音无水印视频链接,视频文案,douyin-mcp-server,mcp,claude skill,支持龙虾
The Open-Source Multimodal AI Agent Stack: Connecting Cutting-Edge AI Models and Agent Infra
A MCP for Claude Desktop / Claude Code / Windsurf / Cursor to build n8n workflows for you
🔥 Official Firecrawl MCP Server - Adds powerful web scraping and search to Cursor, Claude and any other LLM clients.
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP