evidence:<dataset>:mcp scope. Accounts on an organization license need the explicit entitlement even at Pro; email sales to add one.
The shared pattern
Every dataset follows the same two-tool shape: asearch_* tool that returns slim results, and a get_* tool that returns the full record for one id. Patents add a third tool for forward citations.
Search results are deliberately trimmed to save model context. Call the get_* tool only when you need a field the search result doesn’t already carry — the per-dataset sections below say what each one adds.
All search tools share the same parameters:
This is the Evidence query syntax, not the Search syntax.
search_literature uses named parameters and limit/offset; Evidence tools use q/f and page numbers. See Evidence overview.search_510k_summaries, search_mhra, search_faers, and search_drugs — sort by relevance only and accept no s or sortDir.
Date range filters
Datasets with date fields accept two equivalent notations. Note there is no space before thegte/lt suffix.
YYYY-MM-DD, ISO datetimes, and epoch milliseconds.
Patents
search_patents returns patent families with titles, abstracts, inventors, assignees, classifications, filing status, and citation counts.
Sort fields: _relevance, forwardCitationCount, familySize, patents.publications.pubRef.date, patents.appRef.filingDate.
f: patents.assignees.name, patents.inventors.name, patents.filingStatus (application or grant), and patents.languages (e.g. en).
Each result’s id, such as US7732489, is what get_patent and get_patent_citations take.
get_patent returns the full record for one patent, including everything it cites:
publications: every publication in the record, with grant and application numbers, dates, and links.citedPatents: the earlier patents it cites (prior art), each with publication number and date.citedLiterature: the scientific papers and other non-patent literature it cites, as the patent’s reference text, with adoiwhen the reference contains one. Pass those DOIs tosearch_literatureto read the papers.
get_patent_citations finds later patents that cite this one (forward citations). It takes id and an optional page p, and returns citing patent families with the same fields as search_patents.
Citations are matched against the first 10 publications in the patent’s record; publicationsSearched and publicationsInRecord show whether any were left out. Results are deduplicated by family within a page, so totalCitingRecords usually exceeds the number of distinct families. Page with p = 2, 3, and so on. The response has no end-of-results flag: a page that returns no results is the end. A family can appear on more than one page, so a page of only already-seen families usually means you’re near the end, but doesn’t guarantee it. failedLookups counts publications whose search failed.
Clinical trials
search_clinical_trials covers ClinicalTrials.gov. Results carry nctId, title, briefDescription, phase, sponsors, facilities, conditions, interventions, tags, startDate, completedDate, and publicationCount. Matched fields come back wrapped in <strong> tags.
Sort fields: _relevance, dates.startDate, dates.completedDate, dates.lastUpdatedDate.
get_clinical_trial takes an NCT id such as NCT02986230 and adds the full description, study design, enrollment, primary and secondary outcomes, complete eligibility criteria, reportedEvents (posted adverse events), principal investigator, contacts, citations, and resultsUrl. Use it for deeper analysis or patient-trial matching.
Grants
search_grants spans NIH RePORTER, NSF, SBIR/STTR, Wellcome, EU funders, and more. Results carry id, title, an abstract preview, agency, organization, piName, country, award dates, awardAmount, tags, groupSlug, and grantsInGroup.
Sort fields: _relevance, awardStartDate, awardCloseDate, awardNoticeDate, awardAmount, employeeCount.
get_grant takes an id such as 5201339, nsf.0646294, or wellcome.214402.Z.18.Z. Call it for two things only:
- The full
abstract. Search returns a ~300-character highlighted preview; the full text usually runs 1–3 KB. - Source-specific identifiers absent from search:
awardYear,agencyTrackingNumber,contract,nihProgramCode,nihrApplicationId.
Device clearances
The 510(k) dataset splits across two pairs of tools, and the distinction matters.search_device510k returns structured clearance metadata: kNumber, title, summaryText, device info (name, class, product code, clearance type, regulation number), decision info (code, description, date, committee), applicant details, and tags.
Sort fields: _relevance, device.device_class, device.date_received, decision.decision_date.
get_device510k takes a K number such as K210674 and adds the complete summaryText, full applicant details (address, contact, country), registration info (FEI and registration numbers), and the full decision object.
search_510k_summaries searches OCR’d full text from 510(k) summary PDFs and returns page-level snippets. Reach for it when the question is about submission content: test results, performance data, biocompatibility, substantial equivalence reasoning, indications for use, predicate comparisons, sterilization methods, software descriptions, or bench testing.
get_510k_summary takes a document id — also a K number, e.g. K192757 — and returns the full extracted text organized by page.
Adverse events
Two separate FDA systems: MAUDE for devices, FAERS for drugs.MAUDE (devices)
search_maude returns device adverse-event reports with id, title, reportNumber, eventType, adverseEventFlag, productProblems, device info, patientProblems, dates, and narrative snippets.
Facet fields for f: event_type, device.device_class, device.manufacturer_d_name, device.device_report_product_code, device.regulation_number, report_source_code, product_problems, patient.problems. Date fields: date_received, date_report.
Sort fields: _relevance, date_received, date_report.
get_maude_report takes a report id such as 17343805 and returns the full MDR narrative entries with their text type codes, reporter occupation, health professional flag, device medical specialty and availability, and patient treatment. Search truncates these narratives; the detail tool doesn’t.
FAERS (drugs)
search_faers returns drug adverse-event and medication-error reports. Each links suspect and concomitant drugs to observed patient reactions as MedDRA preferred terms. Results carry safetyReportId, title, reportType, serious, seriousnessType, receiveDate, occurCountry, reactions, and drugs.
Facet fields for f: drug.medicinalproduct, drug.brand_name, drug.generic_name, drug.substance_name, drug.manufacturer_name, drug.drugindication, drug.pharm_class_epc, drug.pharm_class_moa, reaction.reactionmeddrapt, reaction.reactionoutcome, event.reporttype, event.seriousness_type, event.occurcountry, event.patientsex, primarysource.qualification, primarysource.reportercountry. Date fields: event.receivedate, event.receiptdate, drug.drugstartdate, drug.drugenddate.
FAERS sorts by relevance only.
get_faers_report takes a safety report id such as 26185565 and adds patient demographics (sex, age group, onset age, weight, death date), report dates, the expedited flag, primarySource (reporter qualification and country), reportDuplicates, and enriched drug records with dosage text, start and end dates, NDC, application number, and pharmacologic class.
Drugs
search_drugs covers FDA Structured Product Labels, the Orange Book, and Drugs@FDA. Each result bundles an FDA application — approved products, applicant, approval dates, marketing status — with its Structured Product Label.
Facet fields for f include labels.brand_name, labels.generic_name, labels.substance_name, labels.manufacturer_name, labels.product_type, labels.route, labels.pharm_class_epc, labels.pharm_class_moa, labels.pharm_class_cs, labels.rxcui, labels.unii, labels.product_ndc, labels.package_ndc, application.application_number, application.sponsor_name, application.products.marketing_status, application.products.dosage_form, application.products.route, application.products.te_code, tags, and categories. Date fields: labels.effective_time, application.products.approval_date.
Sorted by relevance only.
get_drug takes the record UUID from search results, e.g. 4dd865ec-8889-49ac-8c8f-4438875937ac. It adds every approved product (product number, applicant, approval date, dosage form, route, active ingredients, TE code) and the main label sections: indications and usage, dosage and administration, contraindications, boxed warning, warnings and cautions, adverse reactions, drug interactions, use in specific populations, pharmacology, clinical studies, how supplied, overdosage, and description.
Safety alerts
search_mhra covers UK MHRA drug safety alerts, medical device alerts, field safety notices, and regulatory publications, searching headlines, descriptions, and page-level document content. Results carry id, headline, description, tags, categories, date, and matching content snippets.
Facet fields for f: ontology.tags, ontology.categories, domain. Date fields: attachments.file.createdAt, attachments.file.modifiedAt. Sorted by relevance only.
get_mhra_alert takes a document id from search results and returns the complete alert: headline, description, the full article body (contentHtml), tags, and per-page content.
Related
- Evidence API guides: the same datasets over HTTP, with per-dataset filter references
- Tool reference: the full catalog and access requirements
- Errors and rate limits: retry guidance
