
MeltFlex AI
io.github.MeltFlexDevs由 GitHub 帳號 MeltFlexDevs 發布v0.6.0更新於 Nov 7, 2026
AI interior design — restyle rooms, place furniture, and virtually stage spaces from a photo.
概覽
MeltFlex AI 讓助理依照片重新設計房間、擺放家具,並產生 3D 模型、導覽影片與可探索的 3D 場景。
- 可以做什麼
- 依一張照片產生室內外設計圖,涵蓋 16 種模式(換風格、虛擬佈置、格局優化、廚房、浴室、外牆、花園、地板、牆面、樓梯、門窗、局部編輯)
- 用最多 10 張參考圖把指定的真實家具產品放進房間照片
- 依房間圖片產生電影感導覽影片
- 把家具照片轉成 GLB 3D 模型、把 2D 平面圖轉成 3D 模型或算圖、把房間照片轉成可探索的 3D 場景
- 查詢帳戶剩餘點數
- 適用情境
- 適合希望助理依現有照片產出逼真的室內外改造效果、虛擬佈置,或 3D 素材與導覽影片,且已擁有 MeltFlex 付費訂閱的使用者。
- 執行需求
- 需要 MeltFlex 帳戶、有效的付費訂閱,以及透過 MELTFLEX_API_KEY 提供的 API 金鑰(mf_sk_...);也可用該套件的登入指令在瀏覽器中授權,金鑰會存到本機設定檔。可作為遠端端點使用,也可透過 npx 執行本機 npm 套件 meltflex-mcp,本機方式需要 Node.js。需要網路連線。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 MeltFlex AI,將其新增到工作區。
- 連接 所需的 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
- 為需要使用其工具的對話啟用該服務。
Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
交給代理程式安裝
把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。
閱讀 https://sourceweft.com/mcp/io.github.MeltFlexDevs%2Fmeltflex/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。其他 MCP 客戶端
把它新增到你客戶端的 mcpServers 設定中。 客戶端會在首次連接時完成 OAuth 登入。
{
"mcpServers": {
"meltflex-ai": {
"type": "http",
"url": "https://www.meltflexai.com/mcp"
}
}
}README
MeltFlex AI Skills
[License: MIT] [Version] [Skills] [MCP tools] [Website]
AI agent skills for photorealistic interior and exterior design and 3D via MeltFlex AI — restyle rooms, place real furniture, swap floors, walls, stairs, doors and windows, redesign kitchens, bathrooms, facades and gardens, and turn 3D drafts into photos, all from a single photo. Then go further: cinematic video walkthroughs, 2D floorplan → 3D model or rendered 3D picture, furniture photo → GLB 3D model, and photo → explorable 3D world. Works with Claude Code, [圖片] Hermes, Cursor, Codex, Cline, and other AI agents that load Markdown-based skills or speak MCP.
Each user authenticates with their own MeltFlex API key; every generation is billed to their account credits. The skills call the public MeltFlex API directly, so there's nothing extra to install.
[Interior design — empty room to a fully furnished, photorealistic redesign]
[Floor restyle — upload a material and it is applied to your room]
[Garden design — a plain patio restyled into a Mediterranean garden]
Interior redesign · floor restyle from a material · garden design — a few of the 16 image modes, all from one photo.
[Furniture photo → GLB 3D model: a blue armchair and a green sofa, each turned into a textured 3D model]
[Room render → explorable 3D world: a Scandinavian living room turned into a walkable Gaussian-splat world]
[2D floorplan → GLB 3D model]
Furniture photo → 3D model · room render → explorable 3D world · 2D floorplan → 3D model. Real outputs of /api/v1/furniture-3d, /api/v1/world and /api/v1/floorplan-to-3d.
Install
Pick one. Each method sets up the skills for your agent.
npx skills — recommended, cross-agent
GitHub CLI (v2.90+)
Claude Code marketplace
Inside Claude Code:
Setup script
Universal fallback:
More options in INSTALL.md. Agent-driven install (paste into your agent): INSTALL_FOR_AGENTS.md. Cline and other MCP-settings agents: llms-install.md.
Authenticate
You need an active subscription (any paid plan) — without one the key won't work. After subscribing at https://www.meltflexai.com/settings, open account settings → API Key under Profile, then Generate API Key (shown once — copy it):
[Settings — click API Key] [Generate API Key]
Then expose it:
If you use the MCP server / CLI, you can skip the copy-paste and sign in from the browser instead: npx -y meltflex-mcp auth login opens https://www.meltflexai.com/cli/authorize, mints a key from your signed-in session and stores it in ~/.meltflex/config.json (chmod 600).
Skills
Image skills are self-contained: a room photo in, a photorealistic redesign out, 10 credits per image (auto-refunded on failure).
Image modes
meltflex-design — 16 modes, each the same POST /api/v1/generate endpoint with a mode-tuned prompt:
Optional fields on every image call: resolution (512 / 1K / 2K), designLevel (lite −2 cr · quick · pro +5 cr), variations (1–3 per call, charged per image), up to 10 referenceImageUrls.
Video & 3D
Same key, five more outputs. The three 3D builds can take minutes; they answer HTTP 202 + a poll URL when they outlive the request, and /meltflex:3d shows how to poll.
Full reference with response bodies and three.js / Spark loader snippets: https://www.meltflexai.com/api.
Quick Reference
Embed on your website
Live on the Kondela furniture shop: shoppers place its own products in a photo of their room, on the shop's own page.
No API key and no backend: build an embed in the Website Embed Builder (your domain, logo, colour, language, and for shops your product feed), then paste one iframe. Visitors design their room on your page without an account; each design uses 10 credits from your account. Any account gets the code and 10 free designs; after that it runs on an Enterprise plan.
On a shop's product page, add ?product=<product page address> and the tool opens with that product. /meltflex:embed installs the code in your project (HTML, React / Next.js, Vue, Shopify, WordPress), sets the frame-src rule if your site has a Content Security Policy, and walks through a blank frame. Guide: https://www.meltflexai.com/api#embed.
Use with the MCP server
Agents that prefer tool calls over scripted API calls can run the meltflex-mcp server (npm, stdio). It exposes six tools — generate_interior, generate_video, floorplan_to_3d, furniture_to_3d, generate_world, check_credits — and handles polling and saving files to disk.
Any MCP client works with the same shape (command: npx, args: ["-y", "meltflex-mcp"], optional env.MELTFLEX_API_KEY). The same package is a terminal CLI: meltflex generate | video | floorplan | furniture | world | credits. A Claude Desktop extension bundle lives in mcpb/.
[圖片] Hermes
Hermes (Nous Research) speaks MCP, so it runs MeltFlex through the same server.
From the catalog (once merged into the Hermes catalog):
Manually — sign in once, then add the server to ~/.hermes/config.yaml:
Start hermes chat (or run /reload-mcp in a session) to load the six MeltFlex tools. Full guide: https://www.meltflexai.com/mcp.
Going further
- API docs: https://www.meltflexai.com/api
- CLI: https://www.meltflexai.com/cli —
meltflex generatefrom your terminal - MCP server: https://www.meltflexai.com/mcp —
npx -y meltflex-mcp - Cookbook: COOKBOOK.md · Changelog: CHANGELOG.md
Security
The skills never contain a key. Each user supplies their own mf_sk_ key via MELTFLEX_API_KEY; generations spend that user's own credits. Credits, auth, and rate limits are all enforced server-side. Never commit your key — revoke it anytime in settings.
License
MIT — see LICENSE. Contributions welcome — see CONTRIBUTING.md.
來源:README.md,提交 a510b4e
工具
0版本歷史
2- v0.6.0最新Nov 7, 2026
- v0.2.1Sep 16, 2026
評分
已滿足適用檢查項的 85%
該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。
- 近 30 天內可連通必需25 分
- 已宣告認證方式必需10 分
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分