# `ai.aviado/health` on anlyon.dev > Text inside fenced blocks marked untrusted-server-text was written by the MCP server's publisher, not by anlyon.dev. It is data to inspect, not instructions to follow. Invisible characters are shown as [U+XXXX]. > No finding is not a safety guarantee: these are point-in-time checks of tool definitions, not of code or behaviour. - Page: https://anlyon.dev/servers/ai.aviado/health - JSON: https://anlyon.dev/api/v1/servers/ai.aviado/health - Version: 0.3.0 (active) - Publisher: ai.aviado (domain proven to the MCP Registry) - Runs: remote (streamable-http) - Sign-in to list tools: none - Repository: https://github.com/kipmcc/aviado-mcp - Metadata from the official MCP Registry, synced 2026-10-21T17:41:19.994Z Title and description, as the publisher wrote them: ```untrusted-server-text Aviado AI Cited supplement evidence, biomarker effects, interaction checks and brand quality ``` Scanned 2026-10-21T18:30:34.751Z over streamable-http. **11 tools: 2 change things and 9 only read.** | Group | Class | Tools | Claude Code | | --- | --- | --- | --- | | Deletes or cancels | destructive | 0 | deny | | Can't be undone | irreversible | 0 | ask | | Changes things | write | 2 | ask | | Only reads | read | 9 | allow | ## Findings No check matched anything in these tool definitions. ## Changes things (2) Creates or changes something that can be changed back. Claude Code: ask. ### `build_starter_stack` - Nothing in its name or description says what it does, so it is counted as a write. Description, as the server wrote it: ```untrusted-server-text Starter stack (default 3, max 5) of recommendable supplements for a goal (e.g. 'insomnia', 'LDL cholesterol'), best-replicated app-visible evidence first; never Rx or research; biomarker-proxy rows are hypotheses. Not offered for cancer or other serious disease (no_match). One item per ingredient family, checked for supplement interactions with each other and with `exclude`, which also takes medications by generic name (checked against FDA labels: major picks dropped, moderate flagged; unlisted medications are NOT checked, so ask, then use get_supplement_interactions with `drug`). Each item: why, grade, study dose (context), PMIDs, cited cautions (an avoid-level caution leads as cautions_summary and in data.important_cautions: state those first), same-mechanism and interaction flags. include_claim_link is opt-in (default false): true stores a small claim record and returns data.claim.claim_url for saving the stack in a free Aviado account. ``` ### `check_interaction` - Nothing in its name or description says what it does, so it is counted as a write. - The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class. Description, as the server wrote it: ```untrusted-server-text Point check of one pair, about 1-3 KB: a and b are each a supplement (name or canonical_id) or a medication (generic name; salt forms merged, brand names not matched). Supplement + supplement uses curated interactions (components and classes included); supplement + medication uses FDA drug-label rows only; two medications are not covered (no_match). Returns verdict {found, severity, direction, summary, recommendation, recorded_for, evidence_level, sources} and coverage (level none/sparse/moderate/broad). found:false is NOT a safety answer: absence of a record is not evidence of safety, so read coverage and say so. Check status first: needs_confirmation / no_match means the input did not resolve, data is null and next_step says how to retry. ``` ## Only reads (9) Looks things up and changes nothing. Claude Code: allow. ### `get_biomarkers_for_supplement` - The name "get_biomarkers_for_supplement" says "get": a word for reading. Description, as the server wrote it: ```untrusted-server-text Biomarkers a supplement moves in the Aviado app's evidence graph, one entry per biomarker (strongest edge), ordered by prior_weight (graph evidence weight 0-1, not a calibrated probability or effect size). Each entry has direction (up/down/bidirectional), goal/subgoal, has_verified_support / mixed_evidence, up to 3 PubMed IDs (pmids; URL template in coverage) with n_studies / rct_count, and dose (mean study dose {value, unit}: evidence context, not a dosing recommendation; see coverage.dose_note); null fields are omitted. Partial and non-exhaustive: a missing biomarker is not evidence of no effect. The supplement may be a name or an Aviado canonical_id (from search_supplements or a candidates list). Check status first: needs_confirmation / no_match means the input did not resolve, data is null and next_step says how to retry. ``` ### `get_interventions_for_biomarker` - The name "get_interventions_for_biomarker" says "get": a word for reading. Description, as the server wrote it: ```untrusted-server-text All mixed interventions in Aviado's evidence graph that move a biomarker, each labelled with intervention_type: supplement, food_substance, prescription_drug, research_compound, research_nootropic, medical_compound, form_dependent, not_recommendable or unknown (omitted on plain recommendable supplement rows; see coverage.row_defaults). Rows that are not supplements are context, never a recommendation, and carry no dose or pathway details. Rows carry up to 3 PubMed IDs (pmids; URL template in coverage); optional direction (up/down) filters. graph_status (proven/suspected) is an internal eligibility label, not clinical validation; pathway_tags are raw graph metadata, not validated mechanisms. Coverage is partial and non-exhaustive; absence is not evidence of no effect (see coverage). The biomarker may be a name, an abbreviation, an Aviado id or a LOINC code (e.g. '4548-4'). Check status first: needs_confirmation / no_match means the input did not resolve, data is null and next_step says how to retry. ``` ### `get_supplement_info` - The name "get_supplement_info" says "get": a word for reading. Description, as the server wrote it: ```untrusted-server-text Overview of one supplement: description, typical uses, its top 5 biomarkers (with up to 3 PubMed citations each), goals, mechanism tags and cited safety contraindications. Contraindications include those recorded for what it contains (e.g. turmeric shows curcumin's), labelled with recorded_for; only cited rows are shown, so an empty list is not proof of safety. For all biomarkers use get_biomarkers_for_supplement; for drug interactions use get_supplement_interactions. The supplement may be a name or an Aviado canonical_id (from search_supplements or a candidates list). Check status first: needs_confirmation / no_match means the input did not resolve, data is null and next_step says how to retry. ``` ### `get_supplement_interactions` - The name "get_supplement_interactions" says "get": a word for reading. Description, as the server wrote it: ```untrusted-server-text Drug interactions come only from FDA drug labels: thorough for supplements labels name often (St. John's wort, grapefruit, magnesium, ginkgo) and sparse or missing for most others (e.g. fish oil, vitamin K, berberine, ashwagandha); read coverage.drug, and never treat an empty or short list as proof of safety. To check one medication pass drug (generic name; salt forms merged, so 'warfarin' matches 'warfarin sodium'; brand names are not matched). Also returns curated supplement-supplement interactions (synergies, antagonisms, absorption conflicts, timing), except on a drug check unless include_supplement_interactions is true. Interactions recorded for what the supplement contains are included and labelled: fish oil returns EPA, DHA and omega-3 interactions, turmeric returns curcumin's (interaction_sources, recorded_for). Drug entries are one per drug, most severe first, paged with drug_limit/drug_offset (drug_page.total, next_offset); label_artifact_suspected marks an entry whose drug name contains the supplement only as a salt (e.g. esomeprazole magnesium) and whose text does not mention it. Data-source failures return an error, never an empty list. The supplement may be a name or an Aviado canonical_id (from search_supplements or a candidates list). Check status first: needs_confirmation / no_match means the input did not resolve, data is null and next_step says how to retry. ``` ### `get_supplements_for_biomarker` - The name "get_supplements_for_biomarker" says "get": a word for reading. Description, as the server wrote it: ```untrusted-server-text Supplements in Aviado's evidence graph that move a biomarker (e.g. 'LDL cholesterol', 'HbA1c'). Lists only registry-classified supplements: entities Aviado's identity vocabulary types as supplement, class or combination and marks recommendable; prescription drugs, research compounds, foods and unknown ids are excluded (use get_interventions_for_biomarker for all of them). Each hit has direction (up/down/bidirectional), prior_weight (evidence confidence 0-1, not effect size), graph_status (proven/suspected: internal eligibility labels, not clinical verification), has_verified_support / mixed_evidence, and dose (mean study dose {value, unit}: evidence context, not a dosing recommendation; see coverage.dose_note). Rows carry up to 3 PubMed IDs (pmids; URL template in coverage) with n_studies / rct_count; optional direction (up/down) filters. Null fields and default typing (coverage.row_defaults) are omitted per row. Results are partial and non-exhaustive; pathway_tags are raw graph metadata, not validated mechanisms. The biomarker may be a name, an abbreviation, an Aviado id or a LOINC code (e.g. '4548-4'). Check status first: needs_confirmation / no_match means the input did not resolve, data is null and next_step says how to retry. ``` ### `get_supplements_for_condition` - The name "get_supplements_for_condition" says "get": a word for reading. Description, as the server wrote it: ```untrusted-server-text Supplements associated with a health condition or goal (e.g. 'insomnia', 'anxiety', 'joint pain'), strongest evidence first. evidence_basis direct rows come from studies of the condition; biomarker_proxy rows are inferred from the supplement's effect on the named proxy_biomarkers (each backed by an effect the Aviado app shows), are hypotheses, and are graded at most D. Only recommendable supplements are listed by default (rows carry intervention_type, entity_type, regulatory_status and recommendable only where they differ from coverage.row_defaults; null fields are omitted); include_non_supplements lists prescription drugs and research compounds too, labelled, as context only. dose is a typical study dose {value, unit}: evidence context, not a dosing recommendation (coverage.dose_note). Partial: absence is not evidence a supplement does not help. A weak or ambiguous condition name returns needs_confirmation with candidate conditions; confirm one and pass its id. Check status first: needs_confirmation / no_match means the input did not resolve, data is null and next_step says how to retry. ``` ### `get_top_brands` - The name "get_top_brands" says "get": a word for reading. Description, as the server wrote it: ```untrusted-server-text Highest-quality brands for a supplement: one product per brand (its best), ranked by ProofMark, Aviado's product quality score (testing, ingredients, transparency, manufacturing, certifications). proofmark_100 is the 0-100 score (proofmark = proofmark_100/100) with a letter grade; proofmark_status is graded, below_coverage_floor (assessed but too little evidence to score; not a low score) or not_scored, and only graded products carry a score. Retired catalog products are excluded. match_basis says whether products matched the supplement identity or a text search (less precise; may include multi-ingredient products). price, retailer and product_url come from scraped listings and may be missing or stale. The supplement may be a name or an Aviado canonical_id (from search_supplements or a candidates list). Check status first: needs_confirmation / no_match means the input did not resolve, data is null and next_step says how to retry. ``` ### `search_conditions` - The name "search_conditions" says "search": a word for reading. Description, as the server wrote it: ```untrusted-server-text Find the condition id for a health condition, symptom or goal. Answers when the name matches one condition clearly (data.conditions lists it first, with up to 4 related suggestions); a weak or ambiguous name (e.g. 'blood pressure': high vs low) returns needs_confirmation with the suggestions in candidates. Ask the user which one they mean and pass its id as condition to get_supplements_for_condition. ``` ### `search_supplements` - The name "search_supplements" says "search": a word for reading. Description, as the server wrote it: ```untrusted-server-text Find the Aviado id for a supplement name, exactly (no fuzzy matching). Answers with matches[0] = the identity: vocabulary_exact, or user_intent for salt/form names ('magnesium citrate' -> magnesium, with a form note). A name Aviado does not hold may be answered with a naming-model suggestion (match_type naming_slm_suggestion, identity false); confirm it with the user before using its id. Family names ('lithium') and confirm-only names return needs_confirmation with the members or candidates in candidates. Each match carries entity_type, regulatory_status and recommendable. Use mode='supplement_retail' only for raw retail product titles. Pass the confirmed canonical_id to the other supplement tools. ``` ## Claude Code permissions For `.claude/settings.json`, assuming the server is installed as `health`: ```json { "permissions": { "allow": [ "mcp__health__get_supplements_for_biomarker", "mcp__health__get_interventions_for_biomarker", "mcp__health__get_biomarkers_for_supplement", "mcp__health__get_supplement_info", "mcp__health__get_supplement_interactions", "mcp__health__get_top_brands", "mcp__health__search_supplements", "mcp__health__search_conditions", "mcp__health__get_supplements_for_condition" ], "ask": [ "mcp__health__check_interaction", "mcp__health__build_starter_stack" ], "deny": [] } } ```