Context
MCP tool inputSchema/outputSchema is plain JSON Schema and does not restrict @context, so an OO-LD schema works directly as an MCP tool schema with its semantics carried through as grounding. MLCommons Croissant (JSON-LD on schema.org/Dataset) is now MCP-integrated (Oct 2025) and heavily adopted (NeurIPS, HuggingFace, Kaggle, Google). This is a timely M4 outreach demonstrator.
Refs: w3c/yaml-ld#19; modelcontextprotocol#634 (JSON-LD for MCP tools); Croissant + MCP https://mlcommons.org/2025/10/croissant-mcp/
Scope
- Minimal MCP server exposing an OO-LD-annotated tool
inputSchema (native @context), demonstrating the model filling fields by IRI rather than surface order.
- Show the tiered delivery fallback for a strict provider (title/description injection).
- Optional: ingest a Croissant dataset description as an OO-LD schema.
Acceptance
- A runnable demo + short write-up suitable for the Demo Day / a blog post.
Context
MCP tool
inputSchema/outputSchemais plain JSON Schema and does not restrict@context, so an OO-LD schema works directly as an MCP tool schema with its semantics carried through as grounding. MLCommons Croissant (JSON-LD on schema.org/Dataset) is now MCP-integrated (Oct 2025) and heavily adopted (NeurIPS, HuggingFace, Kaggle, Google). This is a timely M4 outreach demonstrator.Refs: w3c/yaml-ld#19; modelcontextprotocol#634 (JSON-LD for MCP tools); Croissant + MCP https://mlcommons.org/2025/10/croissant-mcp/
Scope
inputSchema(native@context), demonstrating the model filling fields by IRI rather than surface order.Acceptance