Create a README.md at the root of the model repository with a YAML front-matter block delimited by --- containing model card metadata fields
Populate required metadata fields: language, license, base_model, datasets, tags, and metrics following the Hugging Face model card spec
Add structured evaluation results in the model-index section with task type, dataset name, and per-metric scores to enable automatic display on the model page
Write free-text sections including Model Description, Intended Uses, Limitations, Training Data, and Bias/Fairness considerations in Markdown below the YAML block
Validate the card with the huggingface_hub library using evaluate.load('model-card') or the Hub's built-in metadata validator before publishing
Known gotchas
The YAML front matter must be syntactically valid — a single quoting or indentation error silently invalidates all metadata fields and the Hub will display the card without structured data
The model-index evaluation results format requires a specific nested structure (results > task > dataset > metrics); flattening it incorrectly causes the Hub to ignore the scores
License identifiers must match SPDX license IDs exactly (e.g., 'apache-2.0', not 'Apache 2.0') — non-standard license strings will display a warning and may affect discoverability
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp