Glbforge

What the glbforge MCP server does

The glbforge MCP server exposes GLBForge’s local 3D asset workflow to MCP clients. Its 28 tools cover mesh inspection, before-and-after comparison, budget analysis, optimization, logo extrusion, Meshy generation, shipping workflows, and related asset operations. The same capabilities are also available through the GLBForge CLI and local Studio interface.

The server is intended for AI-assisted editing and release checks. An agent can inspect a mesh after an edit, compare it with an earlier version, identify topology or transform regressions, and run budget checks before an asset reaches a web application. The project also handles 2D artwork and photographs: logos can be converted deterministically into watertight 3D geometry, while photographic inputs can be sent to Meshy or processed with the subject-lifting matte flow.

How it works

Inspection reports scene and geometry findings such as shell count, open or overlapping geometry, dimensions, up axis, origin placement, and unapplied or mirrored transforms. Findings include rule identifiers and certainty information. The diff workflow compares two files and reports changes to parts, triangles, shells, topology, transforms, nodes, and meshes. Optional visual comparison uses fixed front, side, top, and isometric views with SSIM measurements.

A typical release path is analyze, optimize, then verify. Analysis applies a selected budget profile and returns a report; optimization can weld, simplify, create LODs, and compress content to meet a target. ship combines routing, analysis, optimization, and gating for a GLB, logo, or photograph. The glbforge MCP server returns these operations to an agent as tool results rather than requiring shell commands.

Setup and configuration

Build the repository with pnpm install && pnpm build. The README identifies the MCP package as @glbforge/mcp and the registry identifier as dev.glbforge/glbforge, but it does not provide a client-specific configuration example. The command-line examples invoke the built CLI from packages/cli/dist/index.js.

Usage counting is optional and disabled by default. Set GLBFORGE_USAGE=1 to enable the local counter, or use glbforge usage --enable. Records are written to ~/.config/glbforge/usage.jsonl; GLBFORGE_CONFIG_DIR changes that location. The project states that these records remain local.

Tools and capabilities

• Inspect meshes and scenes, including GLB, GLTF, USDZ, USDA, and USDC inputs.
• Compare revisions and report geometry, topology, transform, and visual changes.
• Analyze assets against mobile, desktop, or product-configurator budgets.
• Optimize geometry and textures, generate LODs, and verify results.
• Extrude logos or other 2D graphics into textured, watertight 3D assets.
• Generate or download models through the Meshy REST API.
• Scaffold a React Three Fiber viewer and export GLB, STL, or USDZ assets.
• Preview and tune photographic subject masks before extrusion.

Limitations and notes

The matte workflow is an inference for photographs, not a guaranteed segmentation result. It reports coverage, pieces, holes, and confidence, and refuses results below its confidence threshold. front is recorded as declared input and is not inferred. GLB files use Y-up by definition, so a requested Z-up value is informational for that format.

Topology processing can be skipped, and non-manifold geometry may cause LOD simplification to use grid vertex clustering instead. Analysis exits non-zero when an asset exceeds its selected budget, which makes it suitable for CI gating. The README does not name specific MCP desktop or editor clients, nor does it state a license.