Optional machine interfaces

The command-line workflow is EyeDataHub's primary interface. Python, JSON, JSON-LD, and Model Context Protocol (MCP) provide secondary views of the same metadata implementation. Agreement among them is a software-conformance check, not independent validation of source records.

Deterministic JSON

eyehub search --modality oct --json
eyehub show airogs --json
eyehub cite airogs --type dataset --format json
eyehub preflight airogs --json

These operations are read-only. Dataset transfer begins only through an explicit non-dry-run eyehub download command.

Python

from eyedatahub.datasets.registry import REGISTRY

dataset = REGISTRY.get_dataset("airogs")
print(dataset.info.access_friction)
print(dataset.info.source_terms)
print(dataset.info.terms_scope)

REGISTRY is retained as an internal compatibility name; the public product is a command-line access tool backed by a versioned catalog snapshot.

Read-only MCP server

python -m eyedatahub.agent.mcp_server

Example client configuration:

{
  "mcpServers": {
    "eyedatahub": {
      "command": "python",
      "args": ["-m", "eyedatahub.agent.mcp_server"],
      "cwd": "/path/to/EyeDataHub"
    }
  }
}

The MCP server supports metadata retrieval and reviewable planning. It cannot transfer data, accept source terms, approve a route, bypass access controls, choose or train a model, or determine legal, ethical, clinical, or scientific suitability. Any software client must preserve unknown values and show the official evidence and dated verification state.

Reproducible interface checks

The test suite covers Python, command-line, and MCP interface conformance, along with manually specified expected-query fixtures. Run it with:

python -m pytest -q