drug_liver_toxicity_lookup
ChatGPT• Looks up a single drug, herbal, or dietary supplement in a reference database of drug-induced liver injury and returns its hepatotoxicity profile - Returns the likelihood score (A = well known cause of clinically apparent liver injury, through E = unlikely cause; E* = suspected but unproven), plus the pattern, mechanism, and management prose - The result is a source document with its link: cite it, with the link, for what you take from it - Use this when asked whether a specific agent causes liver injury, or about its DILI risk, pattern, onset, or management Usage notes: - Pass one agent per call (generic or brand name); for a class question, name the class (e.g. "Beta-Adrenergic Receptor Antagonists") or use drug_liver_toxicity_search - The likelihood score is a causality grade, not a probability — report the letter AND its meaning - The database is a reference, not patient-specific advice; it never replaces clinical judgment - The data comes from a public U.S. government resource. Cite it with its link, and never imply Almanac is affiliated with or endorsed by its publisher
drug_liver_toxicity_search
ChatGPT• Searches a reference database of drug-induced liver injury for monographs relevant to a query and returns the top matches as source documents with their links - Use it for cross-cutting questions ("which antibiotics cause cholestatic injury?") rather than a single named agent — for one agent, prefer drug_liver_toxicity_lookup - Each result carries the agent's likelihood score and the relevant hepatotoxicity prose; cite the ones you rely on, with their links Usage notes: - Keep queries clinical and specific - Report each agent's likelihood grade AND its meaning; the database is a reference, not patient-specific advice - The data comes from a public U.S. government resource. Cite it with its link, and never imply Almanac is affiliated with or endorsed by its publisher
literature_search
ChatGPT• Searches the medical literature and returns the most relevant articles for a query - Use it to ground answers about treatments, guidelines, diagnostics, prognosis, and recent research in published evidence - Results come back as source documents (title, journal, date, abstract, link): read them and cite the articles you rely on, with their links - Results are ranked by keyword relevance, then by evidence level (systematic reviews and RCTs surface first) Usage notes: - This is a keyword search, not a question answerer: write the terms a paper would use ("apixaban warfarin atrial fibrillation chronic kidney disease", not "what's the best blood thinner for afib with bad kidneys") - Include synonyms and MeSH vocabulary for the key concepts ("myocardial infarction" as well as "heart attack"); articles are indexed with MeSH headings - Narrow with filters rather than extra words: publication_types for evidence strength, year_from/year_to for recency, mesh_terms for an exact MeSH heading - Search iteratively: when results are off-target, refine the keywords or filters and search again; issue separate searches for separate questions - Use today's date (in the conversation's context) when relevance depends on recency (e.g. "latest guidelines").
recalculate
ChatGPTRecompute an Almanac calculator card from the values typed into it. Called by the card itself.
search_calculators
ChatGPT• Searches the built-in library of clinical calculators, scores and medical equations (risk scores, severity scores, dosing/physiology formulas, decision rules) - Returns matching calculators with their slug, description, when-to-use note, and the full input contract (input ids, labels, units, allowed options) - Use it whenever a validated score/formula would answer the question better than mental math ("what is her CHA2DS2-VASc", "calculate the corrected sodium", "is this chest pain low risk") — then call use_calculator with the values you can source - Prefer this over computing scores yourself: the library implementation is validated
search_clinical_trials
ChatGPT• Searches a registry of clinical studies and returns matching trials: NCT ID, title, status, phase, sponsor, eligibility criteria, site locations with coordinates and per-site enrollment contacts, central contacts, and the study officials (PI) - Use it when the user asks about clinical trials for a condition, whether a patient might qualify for a trial, or what research is available near a place Search workflow: - Query with condition (+ terms/intervention to narrow). Keep queries clinical and specific. - Finding trials a patient could join: keep the default status (RECRUITING); add NOT_YET_RECRUITING only for soon-to-open studies. - When a place matters, pass location (latitude/longitude of the patient's CITY or the clinic — city-level only, never a street address) so results are limited to trials with a nearby site and sites return nearest-first with distances in miles. - The registry places every site at its CITY's center, so distance_miles is city-to-city: sites in one city share a distance. Say "about N miles", and never rank sites within a city by it. - Results are paginated: next_page_token in the result → pass back as page_token for more. Matching a patient to trials — be rigorous: - Search-mode eligibility text may be TRUNCATED. Before advising on fit for a specific trial, call this tool again with nct_ids (copied EXACTLY from the results — they are case-sensitive identifiers, never typed from memory) to read the full criteria. - Check the INCLUSION criteria against what you know of the patient (age, sex, diagnosis, stage, biomarkers, prior therapies, ECOG/functional status, labs) AND every EXCLUSION criterion (comorbidities, medications, prior treatments, organ function). One failed criterion rules a trial out. - Classify honestly: likely eligible / possibly eligible (list what you could not verify) / not eligible (name the failing criterion). NEVER present eligibility as certain — final screening is always done by the trial site. - Watch the two status levels: a RECRUITING trial can have individual sites that are not yet recruiting or closed — check the per-site status before pointing a patient at a location. - Note the phase and study type when discussing what participation would mean. - Give the NCT ID, the registry link, the nearest recruiting site, and who to contact about enrolling — prefer that site's own contact (name/phone/email), falling back to the central contact. - Trial statuses change; mention that the registry entry is the source of truth. - Trial data comes from a public registry. Cite the registry link with each trial, and never imply Almanac is affiliated with or endorsed by its publisher.
use_calculator
ChatGPT• Runs a clinical calculator from the built-in library and returns its result and interpretation; where the host renders MCP Apps, it also shows the user an interactive calculator card they can adjust - Pass values for every input you can source from the conversation; ask the user for the rest. NEVER guess or invent clinical values - The tool_result gives you the computed result and interpretation — reason about it and weave it into your reply (with the key caveats), rather than reading the calculator back to the user - Get the slug and input contract from search_calculators first
