Skip to main content
Glama

@lunaar/mcp-server

Model Context Protocol server for the Lunaar Public API. Lets Claude, Cursor, Windsurf and any other MCP-compatible host use Lunaar's AI building blocks — sketch-to-render, virtual try-on, jewelry / glasses compositing, body measurement extraction, image-to-3D, AR viewer publishing, upscaling — as native tools, all called by name from chat.

Quick install

You'll need an API key. Sign up at platform.lunaarvision.com → API Keys → New key and copy the lk_live_… value once.

Claude Desktop / Claude Code

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) — or the equivalent on Windows / Linux:

{
  "mcpServers": {
    "lunaar": {
      "command": "npx",
      "args": ["-y", "@lunaar/mcp-server"],
      "env": {
        "LUNAAR_API_KEY": "lk_live_REPLACE_ME"
      }
    }
  }
}

Cursor

~/.cursor/mcp.json (or Settings → Features → Model Context Protocol):

{
  "mcpServers": {
    "lunaar": {
      "command": "npx",
      "args": ["-y", "@lunaar/mcp-server"],
      "env": { "LUNAAR_API_KEY": "lk_live_REPLACE_ME" }
    }
  }
}

Windsurf

~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "lunaar": {
      "command": "npx",
      "args": ["-y", "@lunaar/mcp-server"],
      "env": { "LUNAAR_API_KEY": "lk_live_REPLACE_ME" }
    }
  }
}

Restart the host after editing config. The Lunaar tools should appear in the tool picker / /mcp list.

Related MCP server: Photo AI Studio MCP Server

Tools exposed

Tool

Use it for

Cost

lunaar_sketch_to_render

Convert an interior / exterior / kitchen sketch to a photoreal render.

15 credits

lunaar_studio_tryon

Virtual try-on. Modes: single_item, combo, layer, summer.

10-15 credits

lunaar_studio_poses

Generate alternate-pose variants (Front / Side45 / Side90 / Back) of a previously rendered Studio item.

10 / pose

lunaar_jewelry_tryon

Composite a jewelry product onto a model or display surface. 4 display modes.

10-15 credits

lunaar_glasses_tryon

Drop a pair of glasses onto a face with anatomy-aware fit.

10 credits

lunaar_body_estimation

Pull anatomical measurements from a single full-body photo.

10 credits

lunaar_image_to_3d

Generate a GLB 3D model + spinning preview video from a product photo.

20 credits

lunaar_model_to_ar

Upload an existing GLB / USDZ and get a public AR-viewer URL.

5 credits

lunaar_upscale

Re-render a previously completed /v1/ai/* operation at higher resolution.

5 credits

lunaar_get_operation

Inspect any operation by id (use this when a long-running tool times out before the AI finishes).

free

lunaar_list_operations

Recent operations for the current API key, with status and credit usage.

free

How long-running operations are handled

Most calls reach Completed within 30-90 seconds and the tool returns the final asset URL inline. For Image-to-3D (~3 minutes) the wrapper polls up to 4 minutes; if the host's tool-execution budget is shorter and the operation hasn't finished, the tool returns the operationId and the model can call lunaar_get_operation to pick the result up later.

Failed runs surface the user-friendly error message (Turkish, by design — same string Lunaar's product UI shows end users) and the credit refund total.

Configuration

Env var

Required

Default

Notes

LUNAAR_API_KEY

yes

—

Your lk_live_… key. Always set via the host config's env block — never commit.

LUNAAR_BASE_URL

no

https://api.lunaarvision.com

Override only for staging or self-hosted gateway environments.

Development

git clone https://github.com/Group-Taiga/lunaar-mcp-server.git
cd lunaar-mcp-server
npm install
npm run build
LUNAAR_API_KEY=lk_live_... node dist/index.js

npm run dev re-compiles on save. To test the binary against an MCP host locally, point the host config at the absolute path of the built dist/index.js.

License

MIT — see LICENSE.

Available Tools

14 tools
lunaar_body_estimationB

Extract anatomical measurements (chest, waist, hip) from a single full-body photo. 10 credits per call.

Inputs: heightCm (required), imagePath (required).

lunaar_food_visualA

Turn a casual food photo into a premium food photograph. Pick one of six moods — 1=studio (white background menu shot), 2=delivery_app (clean delivery platform style), 3=lifestyle (real person dining), 4=kitchen (chef-style kitchen, no human), 5=outdoor_social (outdoor restaurant social-media post with person), 6=dining_social (elegant indoor restaurant ambience with person). The AI analyzes the uploaded meal and adapts the environment to the selected mood — the meal itself is never redesigned. 10 credits per call. Latency 15-30s.

Inputs: imagePath (required), mood, aspectRatio.

lunaar_get_operationA

Inspect any operation by id. Use this when an earlier tool returned a non-terminal status, or to poll a long-running Image-to-3D run.

Inputs: operationId (required).

lunaar_glasses_tryonB

Place a pair of glasses on a model's face with anatomy-aware fit. 10 credits per call.

Inputs: personImagePath (required), glassesImagePath (required), aspectRatio.

lunaar_image_to_3dA

Generate a GLB 3D model + spinning preview video from a single product photo. 10 credits per call. Latency 2-4 minutes — may exceed default polling window; if so, use lunaar_get_operation later.

Inputs: imagePath (required), targetHeightCm.

lunaar_jewelry_tryonB

Composite a jewelry product onto a model or display surface. Box (10 credits), ProductModel (10), CustomSurfaceBox (15), CustomModel (15).

Inputs: displayMode (required), productType (required), jewelryImagePath (required), gender, +2 more.

lunaar_list_operationsA

List recent operations for the current API key. Useful for picking up an operationId you forgot or auditing recent calls.

Inputs: page, type, status, pageSize.

lunaar_model_to_arB

Upload an existing GLB (and optional USDZ + thumbnail) and get a public AR-viewer URL. 5 credits per call. Returns inline (sync).

Inputs: title (required), glbFilePath (required), usdzFilePath, thumbnailImagePath.

lunaar_product_shotA

Turn a casual product photo into a premium e-commerce shot. The AI auto-detects the product category — no category input. Pick one of four modes: studio_white (clean white background, marketplace catalog), concept (premium banner / social media scene), lifestyle (in-use / partial human presence — faces avoided), or product_with_model (social-media-ready visual featuring the product with a human model at realistic real-world scale). 10 credits per call. Latency 15-30s. Refunded on provider failure.

Inputs: mode (required), imagePath (required), aspectRatio.

lunaar_product_shot_variantsA

Generate 1..N angle/detail/scale variants of a previously rendered Product Shot parent. Each variant is a separate generation that consumes 10 credits. Variant types: front_view, three_quarter_front, side_view, back_view, top_view, detail_close_up, open_interior (only for products that physically open — falls back to a 3/4 view when not plausible), scale_reference (adds clean dimension annotations using user-provided width/height/depth in cm).

Inputs: variantTypes (required), parentEntityId (required), depthCm, widthCm, +2 more.

lunaar_sketch_to_renderB

Convert a sketch (interior / exterior / kitchen) to a photoreal render. Returns the final image URL. 15 credits per call. Refunded on provider failure.

Inputs: imagePath (required), accessory, sceneType, timeOfDay, +8 more.

lunaar_studio_posesA

Generate 1-3 alternate-pose variants of a previously rendered Studio item. 10 credits per pose. Failed poses are auto-refunded. Not available for Summer or Kids items (those modes have no real subject to retarget).

Inputs: poseTypes (required), parentEntityId (required).

lunaar_studio_tryonA

Virtual try-on. SingleItem (10 credits): person + 1 garment. Combo (15): person + 2-6 garments. Layer (15): person + 3 stacked garments. Summer (15): swimwear with environment. Kids (15): kidswear catalog — 1..N product images on a fully AI-generated child model (no PersonImage); requires kidsGender, kidsAgeRange, and kidsProductCategory.

Inputs: mode (required), fitType, kidsPose, kidsGender, +24 more.

lunaar_upscaleA

Re-render a previously completed AI operation at higher resolution. 5 credits per call. May return inline (200) or queued (202 + poll).

Inputs: operationId (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 14 tools

    Disambiguation5/5

    Each tool targets a distinct task or product category (jewelry, glasses, garments, product shots, food, 3D models, etc.), with clear boundaries. The operations management tools are separate from generation tools, avoiding confusion.

    Naming Consistency5/5

    All tools follow a consistent 'lunaar_<descriptive_verb_noun>' pattern with snake_case, making the purpose immediately clear. No mixing of conventions.

    Tool Count5/5

    14 tools cover a wide range of image generation and manipulation tasks without being overwhelming. Each tool serves a distinct purpose, and the count feels appropriate for the server's scope.

    Completeness4/5

    The tool surface covers core workflows for try-on, product shots, food visuals, 3D generation, AR, and upscaling. Minor gaps exist (e.g., no general background removal or editing tool), but the set is largely complete for its intended domain.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers