x3d_mcp: A Model Context Protocol Server for AI-Assisted X3D Authoring

Abstract

Large language models (LLMs) are increasingly capable of generating structured technical content, but producing valid Extensible 3D (X3D) scenes is brittle in practice: without grounding in the authoritative specification, LLMs hallucinate node names, misuse field types, and emit broken ROUTE connections that fail XSD validation silently. We present x3d_mcp, an open-source Model Context Protocol (MCP) server that grounds LLM-driven X3D authoring in the ISO/IEC 19775 standard via the X3D Unified Object Model (X3DUOM), the X3D 4.0 and 4.1 XML Schemas, the X3D-Edit tooltip profile, and the type-safe X3DPSAIL x3d.py Python library. The server is the result of merging two independent implementations developed in parallel within the Web3D Consortium AI with X3D Working Group, now merged and maintained as an official Web3D Consortium open-source project. It currently exposes 28 tools and four guided prompts across nine modules that cover the full X3D authoring lifecycle (specification discovery, scene synthesis, editing, animation, encoding conversion, validation, and X3DOM rendering in HTML pages); every artifact is built through a four-level validation pipeline: type-safe construction, JSON-Schema validation of tool I/O, XML Schema validation, and authoring-level semantic checks. We demonstrate the server with four end-to-end workflows, including an iteratively refined animated Solar System rendered in X_ITE and a Doha, Qatar city scene assembled from live Google Maps APIs for the Web3D 2026 conference, showing that specification-grounded MCP authoring delivers valid, semantically consistent X3D where unstructured LLM generation does not.