Skip to content
Book Open access

Technē: A Deterministic Craft Layer for Model Context Protocol Servers, Grounded in LLM-Authored X3D

Oct 2026 · Proceedings of the 31st ACM International Conference on 3D Web Technology · pp. 1-12 · 0 citations · 3 references

Abstract

When a language model builds 3D content by calling tools through the Model Context Protocol (MCP), the JSON Schema ensures each call is well-formed—but not that the result will render correctly. A texture placed in the wrong slot vanishes. A humanoid skeleton routed to the wrong parent field is invisible. The schema passes; the pixels fail. We present Technē, a deterministic MCP proxy for the Web3D x3d_mcp server that catches these failures. Sitting between the model and the real server, Technē intercepts each tool call and (1) repairs sloppy arguments using Postel-style coercion, (2) checks a catalog of documented X3D silent-failure modes, returning prescriptive corrections that name the exact fix, and (3) renders the scene and checks for blank frames—a test no schema validator can perform. An opt-in provenance gate extends the same approach to content honesty. On correct input Technē blocks nothing and adds no model round-trips. The system runs no LLM in the enforcement loop, carries 234 passing unit tests, and is measured against the bare server in a controlled comparison. None of this is specific to X3D: wherever a model authors through a tool surface, the craft that surface assumes can be compiled out of documentation and into the transport, where it holds whether or not the model understood it.

Read PDF

Similar papers

Book Open access Oct 2026

Formal Runtime Verification for Tool-Using LLM Agents: An Offline Same-Benchmark Study on AgentDojo and STAC

Results show that formal temporal monitoring adds value exactly when the trace exposes trustworthy history, and quantify how far current benchmarks are from that point and propose a twelve-field enforcement-ready trace schema.

Nikolaos Kekatos, Stylianos Basagiannis, Marinelio Chintri et al. · 2 citations
Open access Sep 2026

A framework for efficient and secure LLM agency: a case for the GraphQL paradigm

It is argued that GraphQL constitutes a principled, testable alternative to function calling for agentic systems, combining lower cost, stronger safety, and improved cognitive robustness.

Viktor Zhakhalov · 0 citations
Book Open access Sep 2026

Demonstrating SAGE: A Dataflow-Native Framework for Modular, Controllable, and Transparent LLM-Augmented Reasoning

This demo presents SAGE as a pipeline-native runtime system and demonstrates it through an OPC-facing control plane and includes a compact distributed comparison against a LangChain RPC baseline over a shared 15-workload RAG suite, where SAGE nearly doubles full-RAG throughput under matched 8-node settings.

Jun Liu, Shu-Hao Zhang · 0 citations
#artificial intelligence Preprint Sep 2026

The Stochastic Deputy: Structural Tenant Isolation for Tool-Using LLM Agents

A structural defense is presented: remove tenant identity from the Model Context Protocol (MCP) tool schema, bind scope to a verified credential, and enforce it below the agent, resulting in a tenant-isolation argument that depends on enforceable interfaces and credentials rather than model compliance.

Mirza Samad Ahmed Baig, Syeda Anshrah Gillani, Asher Ali et al. · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.