Let AI apps query the evidence layer.

The public MCP endpoint exposes read-only tools over the same verified registry, evidence history, compatibility graph and benchmark protocol used by the website.

Streamable HTTP endpointhttps://openmodelweights.com/mcp

search_models

Search verified models with evidence-aware constraints.

get_model

Fetch the full verified record for one model.

compare_models

Return neutral side-by-side fields for 2–4 model IDs.

find_compatible_models

Filter by runtime, format, precision, context, commercial-use class and weight-memory ceiling.

get_model_history

Read the observed evidence ledger for a model.

get_recent_changes

Read recent registry/repository change events.

get_registry_status

Return model count, schema and verification freshness.

get_benchmark_protocol

Return the reproducible deployment benchmark protocol.

The MCP surface is read-only. It returns evidence and constraints; it does not rank political choices, make legal determinations or turn popularity into model quality.