AppLaunchFlow

MCP设计创作

提供方 未标注

设计创作 HTTP · 远程 OAuth 授权 社区 / 第三方 Claude DesktopClaudeCursorVS Code
查看接入方式

通过经过身份验证的 AppLaunchFlow 账户创建和编辑 App Store 截图、社交媒体图片、宣传视频和动态设备样机。

一句话选型结论

按资料字段整理

AppLaunchFlow 属于设计创作类 MCP 服务器,HTTP · 远程,OAuth 授权,可直接接入 Claude Desktop、Claude、Cursor 等,有可直接复制的远程端点配置。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类设计创作 传输方式HTTP · 远程 鉴权OAuth 授权 部署方式支持远程部署
GitHub 星标 未标注 仓库热度,无公开仓库时显示未标注
工具数 未标注 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

设计创作
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量未标注暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

支持客户端

适用场景

设计

安装与配置

HTTP · 远程 OAuth 授权
连远程 HTTPS 端点,通常为 Streamable HTTP
npx add-mcp 'https://mcp.applaunchflow.com/mcp'

这条命令只是入口,具体启动参数请按官方文档补齐,不要直接照搬到别的服务器。

  1. 1
    准备运行环境

    确认你的客户端支持远程端点,并拿到服务地址。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后随便调一个只读工具,确认能返回结果,再接入正式流程。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
出图与素材

生成配图、图标与素材。

设计稿处理

读取设计稿、批量改尺寸或导出。

内容生成

按主题生成文案与视觉草稿。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。
  • 已适配 Claude Desktop、Claude、Cursor,换客户端不用重新研究配置。

先想清楚的地方

  • 远程端点依赖网络与服务方可用性,注意密钥与调用额度。
  • 使用前要准备好OAuth 授权对应的凭据,并按最小权限配置。
  • 资料没标注工具数量,实际能力要装好后确认。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在设计创作场景里让模型直接调用外部能力的人。
  • 已经在用 Claude Desktop、Claude 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 不方便管理密钥或第三方授权的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
AppLaunchFlow当前 未标注 HTTP · 远程 支持远程部署 有 2026-11-20
Cursor Talk To Figma MCP Da1c7911 40 未标注 未标注 无 2026-05-17
Blender MCP 17 stdio · 本地进程 本地部署 有 2026-05-18
Nova Canvas MCP Server 2 stdio · 本地进程 本地部署 有 2026-05-17
Figma Context MCP 3aaa8d28 未标注 未标注 支持远程部署 无 2026-05-17
kreuzberg 未标注 stdio · 本地进程 本地部署 有 2026-04-12

服务器档案

设计创作
字段来自项目仓库与公开索引,中文说明由本站标注
提供方未标注上游标注的维护方 未标注
分类设计创作本站按上游分类展示
连接方式HTTP · 远程连远程 HTTPS 端点,通常为 Streamable HTTP
鉴权方式OAuth 授权远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · 未标注本地安装时用到的包标识与运行时
部署类型支持远程部署上游给出的部署形态
最近核验2026-11-20上游最后核验时间

详细介绍

21.3 KB
来自项目仓库与上游目录的原文,未做删改

AppLaunchFlow MCP

AppLaunchFlow creates app-store screenshots,
social graphics, and AI or device-only videos. This repository provides
its hosted MCP connector for supported AI assistants.

Website · MCP documentation

Pro allowance and Automation

API and MCP exports require an active Pro subscription. Pro includes 5 export jobs per month, shared across all API keys, projects and MCP clients. One job includes its requested languages and device formats. Annual Pro also receives a monthly allowance. Editor exports keep their existing limits.

Add Automation for €19/month on top of Pro for unlimited API exports. Rate limits, job-size limits and AI generation credits still apply.

Use await client.getApiUsage() (requires exports:read) to check used, remaining, unlimited and resetsAt. The API returns HTTP 402 with code api_export_quota_exhausted when the monthly allowance is exhausted, and HTTP 403 with api_pro_required without Pro. Replaying the same idempotency key does not spend another export. Failed and canceled jobs restore their credit; an uncertain dispatch remains reserved until its outcome is known. Polling and downloads are included.

MCP server for AppLaunchFlow — create App Store & Google Play screenshots with AI.

Official API and CI releases

The MCP and the Node SDK now share the versioned /api/v1 API. Import AppLaunchFlow from applaunchflow/api for project-scoped API key authentication, active-variant editing, asynchronous renders and checksum-verified downloads. Browser OAuth remains the normal way to connect MCP.

Use the CI and Fastlane example. Update captures and localized copy directly in the active variant, then render with projectId, languages and formats. CI keys need asset, design read/write and export scopes. HTTP API reference.

Release ordering: 0.7.0 requires the dashboard v1 API and validating bundler to be enabled before this MCP/SDK release is published or deployed.

Release ordering: 0.9.0 requires the dashboard to expose POST /api/v1/promovideo/create-device-only before this MCP release is published or deployed.

Release ordering: 0.8.4 requires the dashboard widget decoder supporting private
payloadJson catalogs. Deploy that backwards-compatible decoder before the MCP:
screenshot/social layouts are now transported as opaque JSON, matching the promo
picker, while model-visible summaries remain unchanged. Widget metadata fingerprints
also invalidate cached resources when the trusted store-icon CSP changes.

Release ordering: 0.8.5 moves screenshot/social picker display to the existing
read-only browse tools. Deploy dashboard support for discovery results before the
MCP. Preparation still supports progress for long requests, but no longer owns a
widget; browsing reads the prepared catalog without generation or mutation.

Connect

Cursor marketplace plugin

This public repository includes a Cursor plugin manifest in .cursor-plugin/plugin.json
and a root mcp.json targeting the hosted service. No API keys or local server are required.
After installing, authorize AppLaunchFlow through Cursor's MCP settings using your
AppLaunchFlow account. Existing plan limits apply; installing the plugin does not
grant a paid AppLaunchFlow subscription.

Try "List my AppLaunchFlow projects", then "Show the screenshot styles for this
project using its existing screenshots". Choose the style in the interactive picker;
preparing previews does not select a design for you. The same flow supports social
graphics and promo videos. Asset uploads and management are available from the asset browser.

If you already configured AppLaunchFlow manually, disable the duplicate MCP entry
when installing the plugin. Support: support@applaunchflow.com.

Maintainers: validate with npm test, then submit the public repository at
https://cursor.com/marketplace/publish. Cursor reviews marketplace listings separately
from npm and the MCP Registry. The Cursor plugin version tracks its packaging release;
the remotely hosted server can receive independent updates.

AppLaunchFlow is a hosted MCP connector with OAuth 2.1 and PKCE. No API key,
local token, or scoped npm package name is required.

Codex

The shortest setup command configures the hosted connector and opens OAuth:```bash
npx -y applaunchflow connect codex


Useful follow-up commands:```bash
npx -y applaunchflow status
npx -y applaunchflow disconnect

The equivalent native Codex commands are:```bash
codex mcp add applaunchflow --url https://mcp.applaunchflow.com/mcp
codex mcp login applaunchflow
codex mcp get applaunchflow
codex mcp remove applaunchflow


### Claude Code

The same one-command setup is available for Claude Code. It replaces a legacy
AppLaunchFlow entry when necessary, adds the hosted HTTP connector, and opens
OAuth sign-in:```bash
npx -y applaunchflow connect claude

The equivalent native Claude Code commands are:```bash
claude mcp add --transport http applaunchflow https://mcp.applaunchflow.com/mcp
claude mcp login applaunchflow


### ChatGPT```bash
npx -y applaunchflow connect chatgpt

Paste the displayed URL when creating a custom MCP connector in ChatGPT. The
same URL is also available at any time with npx -y applaunchflow url.

Other MCP clients

Use this Streamable HTTP endpoint and enable OAuth when prompted:```text
https://mcp.applaunchflow.com/mcp


### Claude Code plugin

This repository is also a distributable Claude Code plugin. Its
`.claude-plugin/plugin.json` manifest bundles the hosted OAuth connector from
`.mcp.json`, so users do not need to copy a server configuration manually.

To validate or try the plugin directly from a clone:```bash
claude plugin validate . --strict
claude --plugin-dir .

Claude Code starts the hosted connector when the plugin is enabled and opens
the AppLaunchFlow OAuth flow when authentication is required.

Hosted service

Editor knowledge and configuration reference

The MCP connector does not train or replace the user's model. It supplies
server instructions, read/edit tools and an on-demand configuration reference.
Before any edit, the instructions direct the assistant to
get_editing_reference(feature), then to retrieve the relevant named symbols.
This also works in clients that do not expose MCP resource browsing.

Supported features are screenshots, socialGraphics, and promoVideo. The index includes the full persisted types, API/renderer validators,
device/format/music/motion catalogs, units, bounds, defaults, and read-before-save
workflows. Examples: ScreenshotNode for overflow/crop/status bars,
Model3DConfig for laptop lids, MultiPhoneContentSchema for promo slots,
and DeviceCustomMotionSchema for keyed folds and lid motion. Complete
references are also available at applaunchflow://editing/; the existing
layout and video-config resource URLs resolve to these current references.

The checked-in reference is generated from an explicit allowlist of dashboard
contracts. It contains no project data or secrets and has no runtime dependency
on a dashboard checkout. To refresh and verify it against the dashboard release:```bash
APPLAUNCHFLOW_DASHBOARD_PATH=../applaunchflow npm run sync:editor-reference
APPLAUNCHFLOW_DASHBOARD_PATH=../applaunchflow npm run check:editor-reference
npm test


Run the drift check whenever editor contracts change and before publishing the
connector alongside a dashboard release. Passing MCP tests alone does not prove
the deployed dashboard matches the snapshot. UI-only actions (selection, undo,
browser rendering, recording camera input) are not persisted configuration and
must not be advertised as MCP tools. Models can still make mistakes: preserve
unrelated fields, read back after saving and verify visual composition where a
preview is available.

### Device-only videos and migrated mockups

Use `create_device_video` to save a device demo from project screenshots or recordings. Use
`list_source_screenshots` for screenshots, `list_video_recordings` for clips, and
`generate_promo_video_motion` for prompt-driven keyframes. The tool returns a Video variant
and an editor link. Use `list_variants(contentType: "promoVideo")` to find migrated mockups,
then the existing `get_promo_video`, `update_promo_video`, `preview_promo_video`, and
`export_promo_video` tools. The old Mockups tools were retired in 0.9.0.

### Inline screenshot picker

`prepare_screenshot_styles` prepares a catalog, then `browse_templates` displays
its cached V1/V2 picker in MCP Apps-compatible hosts. The agent completes both
steps before waiting for the user's choice. Reopen with the same catalogKey;
no generation or saved variant is needed to browse. Social graphics uses
`prepare_social_graphics_styles` then `browse_social_templates`. There are no
additional render tools. Promo videos keep their existing asynchronous picker.
Variants are created only after explicit selection through the authenticated
tool bridge; all hosts retain the full-gallery link.
See local testing and deployment notes.

The public Streamable HTTP service uses OAuth 2.1 authorization code flow with
PKCE through the AppLaunchFlow dashboard.```bash
npm ci
npm run build
APPLAUNCHFLOW_BASE_URL=https://dashboard.applaunchflow.com \
APPLAUNCHFLOW_MCP_PUBLIC_URL=https://mcp.applaunchflow.com \
PORT=8787 \
npm run start:http

Public endpoints:

  • MCP: https://mcp.applaunchflow.com/mcp
  • Protected resource metadata: https://mcp.applaunchflow.com/.well-known/oauth-protected-resource
  • Health: https://mcp.applaunchflow.com/healthz

Official MCP Registry

AppLaunchFlow is published as io.github.ynnickw/applaunchflow in the official
MCP Registry. The checked-in server.json is the canonical
registry manifest and points clients to the hosted OAuth connector.

Registry publication runs automatically from GitHub Actions when the manifest
changes on main. Keep the manifest version aligned with package.json; the
test suite enforces this before publication.

APPLAUNCHFLOW_MCP_PUBLIC_URL may be either the origin or the full /mcp
URL; both services normalize it to the same canonical resource URL. Set
APPLAUNCHFLOW_MCP_PUBLIC_URL=https://mcp.applaunchflow.com/mcp and
NEXT_PUBLIC_APP_URL=https://dashboard.applaunchflow.com on the dashboard.

The included Dockerfile produces a non-root OCI image for the hosted server.
The MCP host and dashboard must both be served through public HTTPS in
production. Do not expose the Node process directly without a TLS-terminating
platform or reverse proxy.

Personalized style workflow

Screenshot and social-graphics styles are prepared once for the selected app screenshots, then the chosen style is applied from cache. This lets users compare styles without paying for another AI call when they choose one.

For App Store screenshots:

  1. list_source_screenshots
  2. prepare_screenshot_styles with 3-7 ordered paths
  3. browse_templates with the returned generationId/catalogKey and requested deviceType displays the cached picker. Compare V1/V2, then confirm the choice; the picker creates the new variant and opens the editor.

browse_templates can reopen a prepared catalog, and apply_screenshot_style
remains available when an API client supplies a template id directly.

For social graphics:

  1. list_source_screenshots
  2. prepare_social_graphics_styles with 3-7 ordered paths
  3. browse_social_templates with the returned generationId/catalogKey and primaryFormat as format displays the cached picker. The user chooses a template and palette there;
    the choice creates a variant with every format and opens the graphics editor.

browse_social_templates also reopens a prepared catalog, and
apply_social_graphics_style is for an API client or user that supplies a
template id directly. Agents do not call it after preparing styles.

The screenshot result includes phone, tablet, and desktop. The social result includes every supported social, store-listing, and mobile-ad format. New screenshot and social-graphics variants always go through their visual pickers so the user explicitly chooses the template and palette.

Scope, plans and limits

get_capabilities lists, per area, the tools this server exposes and the
editor URL, plus everything that only the dashboard can do (store publishing,
store connections, custom product pages, A/B tests, reviews, app icon, website design
and publishing, export history, collaborators, QR codes, API keys, billing) with a deep link for
each. Pass projectId for project links. get_plan_usage reports the monthly
API export allowance and, with projectId, keyword and competitor limits.

Plan and quota rejections (VARIANT_LIMIT_REACHED, PROJECT_LIMIT_REACHED,
api_pro_required) return error.recovery with retryable: false, guidance
and an upgradeUrl. Only HTTP 429 is marked retryable.

Other tools in this area:

  • Store listing text: get_store_listing, generate_store_listing,
    update_store_listing (merges only the fields passed, enforces store
    character limits), translate_store_listing.
  • Keywords: add_keywords resolves the store and app from the project's linked
    app; remove_keywords, suggest_keywords, add_keyword_competitors,
    remove_keyword_competitors, suggest_keyword_competitors.
  • Variants: create_variant makes an empty variant. The first variant of a
    content type becomes active automatically. Later variants start inactive.
    duplicate_variant copies content and returns an editor URL;
    set_active_variant refuses empty variants; delete_variant needs
    confirm: true.
  • create_project accepts appStore (Apple id or apps.apple.com URL) to import
    name, category, description and icon and link the app.
  • Tools that take generationId also accept projectId; both are the project id.

Releasing

  1. Resync the editor reference against the dashboard commit being released and
    review the diff:
    APPLAUNCHFLOW_DASHBOARD_PATH=../applaunchflow npm run sync:editor-reference
  2. Bump the version in every manifest (see scripts/release-consistency.mjs).
  3. APPLAUNCHFLOW_DASHBOARD_PATH=../applaunchflow npm run release:preflight
    checks the reference against that dashboard and writes
    editor-reference.lock.json for the new version. It refuses when the
    reference is stale.
  4. npm test, commit (including the lock file) and merge.

CI and the npm publish workflow run release-preflight.mjs --verify, which
refuses a version increase without a lock written for exactly that version and
reference content. The dashboard repo runs the opposite check
(npm run check:mcp-editor-reference there), so a dashboard contract change is
flagged before the MCP is resynced.

Development```bash

npm ci
npm run dev
```

Run npm test before publishing or deploying. See
docs/openai-submission.md for the final OpenAI
submission checklist and manual test cases.

Embedded project and asset lists

list_projects shows a compact list with app icons, explicit project selection,
and dashboard links. list_assets shows uploaded screenshots, illustrations,
icons, recordings, backgrounds, fonts, and music/clips using the same browser
component as the dashboard Hub. The Upload button accepts up to 10 files per
batch (25 MB per file), with progress and a refreshed list after completion.
Screenshots use the selected device/platform folder; other uploads use the
selected asset tab. Uploading stores files only: it does not replace the app
icon, apply a design, register a font family, or overwrite existing assets.
Asset deletion remains an explicitly confirmed action in the Hub.

The app-only prepare_asset_upload tool uses the connected account to authorize
one uniquely named storage object. Its signed URL is returned in widget
metadata (with the Cursor compatibility fallback described below). The browser sends the file bytes directly to storage; neither file
contents nor account credentials pass through the model. Refreshing calls the
read-only list_assets tool. No upload is triggered by merely opening the list.

Project selection uses standard MCP Apps messaging, or ChatGPT's native
sendFollowUpMessage bridge when the host does not advertise text messaging.
Hosts with neither retain the exact copy/paste fallback. A failed/ambiguous send
is never automatically retried through another bridge.

Cursor currently drops tool-result _meta before delivering results to embedded
apps. For HTTP requests with a Cursor/ user agent, the connector mirrors only
the seven known widget payloads into structuredContent.widgetDataJson.
All five widgets prefer _meta and accept this structured fallback, including
asset-folder and upload responses. Structured content can be model-visible;
never add arbitrary metadata or account credentials to this allowlist. ChatGPT
and Claude keep their original metadata-only delivery. Test the stripped-metadata
path with the dashboard harness ?picker=projects&scenario=cursor (also supports
assets, screenshots, social, and promo).

For maintainers, regenerate the packaged list views and shared contracts with
npm run sync:dashboard -- /absolute/path/to/dashboard. Add --check to verify
they match without updating the connector artifacts. npm test checks artifact
integrity and contract hashes without requiring another checkout or network.
Existing screenshot, social and promo picker behavior is unchanged.

Widget resource URIs include a fingerprint of their sandbox/CSP metadata as well
as the bundle or UI version. A metadata-only fix therefore gets a new cache key,
including in hosts that cache resource documents for an entire conversation.
Generic MCP Apps metadata leaves sandbox identity to the host; the dedicated
OpenAI domain is declared only through openai/widgetDomain. Known, well-formed
older widget version/digest URIs still resolve to the current public UI payload,
without changing account authorization or expanding CSP origins.

Hosted MCP runtime metrics

Set both APPLAUNCHFLOW_MCP_POSTHOG_KEY (the PostHog project token, not a personal
API key) and APPLAUNCHFLOW_MCP_POSTHOG_HOST (for example,
https://eu.i.posthog.com) for the required hosted $mcp_tool_call events.
Use the ingestion host for the same project that receives dashboard/landing events.
The hosted server refuses to start if either variable is missing or the host is invalid.
Local stdio calls are excluded because they have no verified hosted account identity.

Events use the account ID returned by authenticated token introspection and contain
only the tool name, client family, duration, outcome and fixed error category.
Arguments, outputs, asset URLs, prompts, bearer tokens and project contents are
excluded. Capture runs in the background, is bounded, and never retries a tool or
changes its result when analytics is unavailable. Existing Railway logs remain.
No session or agent-intent grouping is inferred from stateless requests.

Long-running style generation sends progress when the MCP request supplies a
progress token. Clients must request progress and reset their timeout on it;
a fixed client deadline can still end the call. After an uncertain generation,
inspect project/catalog state before retrying. For rendering, reuse the original
idempotency key and poll get_release_render until the job reaches a terminal
status; acceptance alone does not prove the package succeeded.

Interrupted screenshot uploads

upload_screenshots reports each completed file with zero-based sourceIndex
and fileIndex, its original sourceFilename, stored path, and
folderAssigned. Folder assignment runs after uploads in chunks of up to 50
paths, rather than consuming one mutation per file.

An incomplete batch includes recovery.phase, the failed source/file indices,
and recovery.nextStep. Retry only the failed and remaining sources. If only
folder assignment failed, use move_assets on unassigned paths without
uploading again. A pendingUpload or pendingOverwrite means a write's outcome
needs checking before retrying. No signed upload URLs or source URLs are
returned in recovery diagnostics. Quota exhaustion (HTTP 402) is classified as
payment_required / api_4xx in tool telemetry.

Independent phone-format layouts

get_layout includes format_layouts for sizes edited separately from the main phone layout. Editing mobile intentionally preserves those independent designs. To edit Duo Open, read the language and variant first, then call transform_layout with formatId: "ios-duo-inner" and operations (omit layouts). To return that size to the main design, call the same tool with formatId, resetFormat: true, no operations, and the fresh read receipt. save_layout also accepts the complete formatLayouts map: omit it to preserve existing overrides; {} resets all overrides. list_export_formats includes ios.phone.duo-inner and ios.phone.duo-outer; preset exports use the same saved format as the editor. Store-validated Fastlane exports retain the ten-image iOS limit.

App Store creative editing

Product Page Header (app_store_header, 3840×1646) and Search Results Creative
(app_store_search, 1920×1280) use the same graphics tools as social designs.
Prepare/browse/apply styles, then read with get_graphics_format and save the
complete edited layout with save_graphics_format, passing the fresh read receipt.
Text, images, logos, device frames, backgrounds and positions remain editable;
the save returns the existing result-preview widget. Other formats are preserved.
Header preparation accepts 1–7 screenshots; other formats require 3–7.
These tools edit designs inside AppLaunchFlow, not published App Store listings.

常见问题

8 问
答案由本站依据条目字段整理,不引用上游原文
AppLaunchFlow 怎么安装?

在客户端按下面的安装命令接入:npx add-mcp 'https://mcp.applaunchflow.com/mcp'。装完在客户端配置里引用同一个包或端点。

AppLaunchFlow 怎么在 Claude Desktop、Claude、Cursor 里配置?

它的传输方式是HTTP · 远程,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

AppLaunchFlow 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

AppLaunchFlow 需要 API Key 吗?

资料标注需要OAuth 授权认证,使用前准备好对应凭据,并按最小权限配置。

AppLaunchFlow 支持哪些客户端?

资料标注支持 Claude Desktop、Claude、Cursor、VS Code、Codex、Zed;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

AppLaunchFlow 能提供哪些能力?

资料显示能力标签包括 设计。具体工具清单以官方文档为准。

AppLaunchFlow 需要什么运行环境?

属于远程端点,不需要在本地装运行时。

AppLaunchFlow 是本地还是远程?

它是远程端点,连服务地址即可。传输方式为HTTP · 远程。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 2026-11-20。