Model Context Protocol

Clarity MCP

Condition-aware ingredient intelligence for agents. Every answer carries its verdict, its evidence tier, and — honestly — how well-evidenced it is.

Transport Streamable HTTP · JSON-RPC 2.0 at POST /  ·  v0.7.8

Tools

check_ingredient

Verdict, evidence tier, evidence_state & citations for an ingredient — one lens, or several conditions in one call via lenses.

check_stack

Audit a supplement stack: per-item condition verdicts plus every curated interaction within the stack, in one call.

validate_claim

Fact-check a free-text health claim against Clarity's cited sources — supported, contradicted, or unsupported, with the citations that decide it.

scan_barcode

Resolve a product by barcode, flag its ingredients for the lens, and surface any active FDA recall.

score_product

Category-distinct quality scores — food, skincare, and supplements (dose transparency, flagged botanicals) — with match-coverage confidence.

find_alternatives

Safer same-category swaps for a flagged product — quality-score ranked, lens-flag screened, coverage-honest. Verify the pick with scan_barcode.

check_interaction

Curated ingredient-to-ingredient interactions — severity, mechanism, and source — for one ingredient (one-to-many) or a specific pair.

strain_lookup

Cannabis & mushroom species verdicts with tier and PMID references.

check_recall

Consumer-product recall matches by brand and product name, with source URLs and a medical-device scope boundary.

Evidence model — honesty over completeness

Gold primary-source + human-reviewed Silver literature-cited + reviewed Bronze curated, review in progress

When a verified citation is on file, a lens reports evidence_state: cited with that citation; when only an unverified, category-level citation exists it reports referenced. Otherwise the verdict stands on Clarity's curation. The verdict and evidence tier are always there. Every result also carries human_url — a live Clarity page for that exact answer, so agents can hand their user a verifiable source link.

Quick start

curl -s https://mcp.healthai.com \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"check_ingredient",
                 "arguments":{"name":"aged cheddar cheese","lens":"histamine"}}}'

List tools with {"method":"tools/list"}. Condition lenses: breastfeeding, pregnancy, histamine, mcas, rosacea, hs, allergy, fertility, toddler.

Access

Free evaluation, 20 work units/minute and 100/day per anonymous source. JSON-RPC batches are limited to four messages and charged by bounded per-tool work; response rows, unique records, and bytes are capped, and quota/data-service failures fail closed. Anonymous barcode and claim calls are read-only; authenticated callers must explicitly opt into bounded mirror/crosswalk or redacted claim-review writes. A self-serve key adds 100 requests/month; it is not a free scrape bypass. Higher-volume paid API access is available at Clarity API plans; bulk snapshots and commercial MCP access are licensed — hello@healthai.com.