Jetty.io

Validates dataset metadata against the MLCommons Croissant schema specification, checking structural conformance, required fields, and semantic correctness of dataset descriptors. Implements schema-based validation that parses JSON/YAML dataset manifests and reports detailed validation errors with field-level diagnostics, enabling developers to ensure their datasets comply with the Croissant standard before publication or use in ML pipelines.

Generates valid MLCommons Croissant metadata files from high-level dataset descriptors or natural language descriptions, using schema-aware code generation to produce compliant JSON/YAML manifests. The generator maps user-provided dataset properties (name, description, splits, features, licenses) to Croissant schema fields, handling nested structures and semantic relationships, and can be invoked via MCP to enable LLM agents to create dataset metadata programmatically.

Implements a Model Context Protocol (MCP) server that exposes dataset metadata operations (validation, generation, querying) as callable tools for LLM agents and applications. The server handles MCP protocol negotiation, tool registration, request/response serialization, and maintains a stateless interface for composable dataset workflows, enabling agents to chain metadata operations without direct file system access.

Enables querying and inspecting Croissant dataset metadata files to extract specific fields, validate completeness, and provide structured summaries of dataset properties. Implements path-based field access (e.g., querying splits, features, licenses) with support for filtering and aggregation, allowing developers and agents to programmatically inspect dataset metadata without parsing raw JSON/YAML.

Processes multiple dataset metadata files in batch, applying validation, generation, or transformation operations across a collection of datasets. Implements parallel or sequential processing with aggregated reporting, error handling per-dataset, and summary statistics, enabling teams to validate or migrate large dataset catalogs without manual per-file operations.