Meshy MCP Server

Model Context Protocol (MCP) server for the Meshy AI 3D generation platform. Enables AI agents to create, manage, and download 3D models, textures, images, rigged characters, and animations through natural conversation.

24 tools covering the full Meshy API:

• Text to 3D: Generate 3D models from text descriptions (preview + refine pipeline)
• Image to 3D: Convert single or multiple images into 3D models
• Meshy 7.1 (v0.6.0): ai_model: "meshy-7.1" (and latest) on text-to-3d, image-to-3d and multi-image-to-3d. geometry_resolution: "standard" | "2k" | "4k" picks the geometry pass (+5 credits for 2k/4k; multi-image supports standard/2k). ultra_mode is deprecated
• Meshy 6 Lite (v0.6.0): ai_model: "meshy-6-lite" — fast, 5-credit mesh (15 textured, 2K textures only) on every generation endpoint
• Smart Topology: model_type: "smart-topology" with ai_model: "meshy-t2" on text-to-3d (new in v0.6.0) and image-to-3d. Clean, part-separated triangle mesh generated directly at 100–15,000 faces (default 4,000) for 5 credits of mesh
• Multi-view texture on multi-image (v0.6.0): texture_image_urls — 1–4 views of the same object drive the texture (Meshy 7.1)
• 8K Textures (v0.5.0): texture_resolution: "2k" | "4k" | "8k" on image-to-3d, multi-image-to-3d, text-to-3d refine and retexture (8K costs 15 credits vs 10). Replaces the now-deprecated hd_texture flag
• Multi-view Retexture (v0.5.0): multiview_image_urls — 1–4 ordered views of the same object drive the texture instead of a single style reference (requires Meshy 7)
• Auto-Rigging & Animation: Add skeletons and animations to humanoid characters
• Creative Lab: One tool — a photo → a finished, print-ready product. 7 products (v0.5.0): figure, lamp, keychain, fridge-magnet, vinyl-figure, brick-figure and keycap. Runs prototype→build end-to-end and returns only the final 3D model
• Format & Size Utilities: convert (format conversion incl. 3MF, 1 credit), resize (real-world dimensions, 1 credit), uv_unwrap (clean UV layout for external texturing, 5 credits)
• 2D Image Models: text_to_image / image_to_image support nano-banana, nano-banana-2, nano-banana-pro, gpt-image-2, gpt-image-2-5-flare and gpt-image-2-5-sunburst
• 3D Printability Suite:
  • analyze_printability — free FDM check (watertight, volume, holes, non-manifold edges, degenerate faces)
  • repair_printability — 10-credit topology repair, now also accepting .fbx / .gltf input (v0.5.0)
  • process_multicolor — 10-credit multi-color 3MF for AMS/MMU printers
• Slicer Integration: Auto-detect 7 installed slicers (OrcaSlicer, Bambu, Creality, Elegoo, Anycubic, PrusaSlicer, Cura) and return launch commands the agent can execute
• Smart File Organization: Auto-saves to meshy_output/ with project folders, metadata, and history tracking
• Built-in Workflow Intelligence: Server instructions guide the agent through correct tool chains for each use case

ai_model is not the same set on every endpoint, and latest does not resolve to the same model everywhere (text, image, multi-image, retexture):

Legacy ids still accepted: meshy-7 on generation (deprecated, billed like meshy-7.1) and single-image meshy-t1. meshy-5 is retired and no longer accepted.

Generation cost (pricing):

Text-to-3D preview is billed mesh-only; refine adds 10 (15 at 8K). geometry_resolution: "2k" or "4k" (Ultra pass, Meshy 7.1 / latest) adds +5; multi-image supports "2k" only. ultra_mode is deprecated and maps to "2k". model_type: "lowpoly" retires on 2026-10-30: use smart-topology + meshy-t2.

Smart Topology (meshy-t2) generates directly at target_polycount (100–15,000 faces, default 4,000) and skips remeshing. For standard models, target_polycount needs should_remesh: true and is overridden by decimation_mode.

Retexture is 10 credits (15 at 8K). Creative Lab is 36 credits for every product (6 + 30; lamp is 30 + 6) except keycap, which is 62 (12 + 50). Other tools: remesh 5, rig 5, animate 3, convert 1, resize 1, uv-unwrap 5, analyze-printability free, repair-printability 10, multicolor 10.

• Node.js >= 18
• A Meshy API key (get one here — requires Pro plan or above)

Pick whichever fits your workflow — they all produce the same config.

add-mcp auto-detects every AI client on your machine (Cursor, Claude Code, Claude Desktop, Windsurf, Codex, VS Code, Cline, …) and writes the right config to each:

npx add-mcp @meshy-ai/meshy-mcp-server --env MESHY_API_KEY=msy_YOUR_API_KEY

After it finishes, jump to Activate for your client.

Already chatting with Cursor / Claude Code / Codex? Paste this prompt:

Install the Meshy MCP server for me. Docs: https://github.com/meshy-dev/meshy-mcp-server
Use this env var: MESHY_API_KEY=msy_YOUR_API_KEY

The agent will run add-mcp (or write mcp.json directly) and tell you when it's ready. You'll still need the Activate step for your client.

{
  "mcpServers": {
    "meshy": {
      "command": "npx",
      "args": ["-y", "@meshy-ai/meshy-mcp-server"],
      "env": { "MESHY_API_KEY": "msy_YOUR_API_KEY" }
    }
  }
}

claude mcp add-json meshy '{"command":"npx","args":["-y","@meshy-ai/meshy-mcp-server"],"env":{"MESHY_API_KEY":"msy_YOUR_API_KEY"}}'

Most clients auto-load the new server, but Cursor and VS Code require a manual toggle:

• MESHY_API_KEY environment variable is required — returned by every tool when the key didn't reach the server. Make sure it sits inside an "env": {...} block in your mcp.json, not in args. Restart the server after fixing it.
• spawn npx ENOENT (Windows) — wrap with cmd /c (see Cursor block above).
• error: unknown option '-y' (Claude Code on Windows) — use claude mcp add-json instead of claude mcp add … -- npx -y ….
• Cursor doesn't list meshy — make sure mcp.json is valid JSON (no trailing commas), then fully restart Cursor.
• Authentication failed: the Meshy API rejected MESHY_API_KEY (401) — the key is mistyped, revoked or expired. Regenerate at https://www.meshy.ai/settings/api, update your config and restart the server.

# Clone and install
git clone https://github.com/meshy-dev/meshy-mcp-server.git
cd meshy-mcp-server
npm ci

# Development with hot reload
npm run dev

# Build
npm run build

# Type check
npm run lint

# Build and run offline mocked-request tests (no API key or credits required)
npm test

# Run
npm start

See setup, updates, and rollback for running a reviewed local build, pinning package versions, testing model routing without paid API calls, and reconnecting a client safely.

For remote access, run in HTTP mode:

TRANSPORT=http PORT=3000 npm start

Endpoints:

• POST /mcp — MCP protocol endpoint
• GET /health — Health check

MIT