This is a collection of Rust CLI binaries designed for data ingestion workflows, NOT an MCP server. The repository contains no MCP protocol implementation, no tool registration mechanism, no schema definitions, and no input/output documentation. All 18 'tools' are inferred from directory names and Cargo.toml files in the dataload folder. There is no evidence of JSON Schema input definitions, parameter descriptions, output structures, or error handling designed for LLM consumption. The tools are standalone CLI utilities (grebi_ingest_*, grebi_transform_*, grebi_parquet2jsonl, grebi_rdf2jsonl, grebi_nodes2edges, grebi_normalise_prefixes) intended to be orchestrated by Nextflow workflows, not by MCP clients or LLMs. No tool definitions, schemas, or descriptions are visible in the source code provided.

• PREREQUISITE: Implement a full MCP server with JSON-RPC protocol support, tool registration via the 'tools/list' method, and request/response handling. See https://spec.modelcontextprotocol.io/latest/basic/server-capabilities/
• Define input schemas for each tool as JSON Schema with 'type', 'properties', 'required', and per-parameter 'description' fields. Example: grebi_ingest_sqlite should define parameters for input_path (string, required), output_path (string, required), table_filter (optional), and format_options (optional).
• Expand tool descriptions to 50 - 200 characters. Include WHAT each tool does, WHEN to call it (e.g., 'after ingesting raw data, normalize prefixes'), and WHAT it produces (e.g., 'outputs GrEBI JSONL with normalized CURIE prefixes'). Reference https://arcade.dev/patterns/tool-description
• Document output schemas. Specify which fields each tool returns (e.g., grebi_ingest_biomappings returns {success: boolean, records_ingested: integer, output_file: string, errors: []}). This enables downstream tool chaining.
• Add parameter descriptions for each input. Example for grebi_ingest_sqlite: 'input_path: Path to SQLite .db file (required). output_path: Directory to write GrEBI JSONL output. table_filter: Comma-separated list of table names to ingest (optional; if omitted, all tables are processed).'
• Define enums for constrained parameters. If a tool accepts a 'format' parameter, list valid values: format (enum: jsonl, parquet, turtle, rdf_xml).
• Add error handling guidance. Document common failure modes and suggest recovery steps. Example: 'If file not found, check that input_path is absolute or relative to the CWD. If schema parsing fails, validate the input format matches the expected structure.'