{"name":"OralInsulin Network — Evidence Layer","description":"Read-only, category-neutral evidence layer for oral insulin — any oral (including oromucosal) insulin that works, from any source; no single product. Serves sourced market and cost records, the eight evidence standards any candidate has to meet, scenario models, and public evidence on candidates as it is found. This surface makes no efficacy, safety, or regulatory claim of its own and returns every record with its source and confidence label. Not medical, regulatory, investment, or valuation advice.","url":"https://oralinsulin.org/mcp","provider":{"organization":"OralInsulin Network","url":"https://oralinsulin.org"},"version":"1.1.0","protocolVersion":"1.0","preferredTransport":"JSONRPC","capabilities":{"streaming":false,"pushNotifications":false,"stateTransitionHistory":false},"defaultInputModes":["application/json"],"defaultOutputModes":["application/json"],"skills":[{"id":"list_claims","name":"List published claim records","description":"Retrieve every claim record the site publishes. Each record carries its metric, value, unit, scope, evidence type, source attribution, confidence label where published, and supersession chain. Records are market and route-science claims only; candidate evidence is reported with its source, and no efficacy claim is available from any skill.","tags":["evidence","claims","diligence"]},{"id":"get_claim","name":"Retrieve one claim record by claim_id","description":"Fetch a single claim record (e.g. INS-TAM-2026-001) with its full published fields, source attribution, related evidence gate, and an explicit list of any record-shape fields the site does not publish for it.","tags":["evidence","claims","diligence"]},{"id":"get_gate_status","name":"Read evidence-gate status","description":"Return the eight evidence gates (product identity, human PK/PD, clinical efficacy, safety, CMC, regulatory, intellectual property, commercial), each with its evidence standard and current status. These are standards any oral insulin candidate has to meet. An agent asking whether a candidate works receives the standard and what the sourced public record shows — not a conclusion.","tags":["evidence","gates","status"]},{"id":"run_scenario","name":"Run the published commercial scenario model","description":"Compute Model A exactly as the site does: peak annual net sales = $19.4B core TAM × clinically addressable fraction × peak adoption × relative net price, with inputs bounded to the site's published ranges. Output is labeled a scenario output — not a forecast or valuation, not probability-adjusted, and conditional on a candidate proving it works. Also returns the published Conservative/Base/Upside set.","tags":["scenario","model","commercial"]},{"id":"list_regions","name":"List estimation regions and published-data coverage","description":"Regions covered by the daily systemwide injectable-insulin cost-estimation pipeline, each with its tier-1 published-data status and declared-unpublished-field count. Coverage follows evidence: regions without published data report fields_not_published, never borrowed estimates.","tags":["estimation","regions","coverage"]},{"id":"get_cost_estimate","name":"Systemwide cost estimate for one region","description":"Two-tier estimate of what injectable insulin costs a health system: tier 1 direct product acquisition cost from published benchmarks (US: CMS NADAC); tier 2 the injectable route's documented system-cost pool (administration, complications, adherence burden), itemized with per-figure citations. Oral-side numbers are always scenario-labeled — no oral insulin is marketed, so no oral insulin price exists.","tags":["estimation","cost","systemwide"]},{"id":"get_global_delta","name":"Global fan-in across regions","description":"Merged global view: published context anchors (IDF Atlas), quantified scenario-savings components with the stated combination rule (tiers reported separately, never silently summed), per-region coverage, and the default EV analysis. Scenario outputs, not forecasts.","tags":["estimation","global","scenario"]},{"id":"get_methodology","name":"Estimation methodology (global or per region)","description":"The two-tier definition, combination rule, scenario parameters with provenance, analytics engine, and per-region sources, validation results, and fields_not_published declarations.","tags":["methodology","provenance","estimation"]},{"id":"get_run_provenance","name":"Full provenance for one pipeline run","description":"Inputs, sources with retrieval dates, per-region worker timings and rows scanned (telemetry), and outputs for a given run_id. Unknown run ids return the valid-id list.","tags":["provenance","telemetry","runs"]},{"id":"get_ev_analysis","name":"Expected-value analysis (founder framing, labeled)","description":"Founder-direction EV layer, labeled as such: expected systemwide savings = p(success) × scenario savings, computed over quantified published components only (unquantified components contribute zero). threshold_p defaults to the founder guess of 0.19. A scenario/EV analysis — never a forecast, valuation, or probability evidence about any candidate.","tags":["scenario","expected-value","estimation"]},{"id":"get_changelog","name":"What changed in the estimates, and why","description":"The estimate ledger's change log: per pipeline run, which published figures changed (from, to, delta) and the candidate cause — input_changed, model_changed, assumption_changed, first_publication. A figure that moves with no recorded cause fails the run and is never published. Append-only and hash-chained.","tags":["ledger","history","provenance"]},{"id":"get_estimate_history","name":"Full history of one published figure","description":"Every recorded value of one figure across runs, oldest first, flagging where it changed and the candidate cause. Search figure paths by substring; unknown paths return suggestions.","tags":["ledger","history","estimation"]},{"id":"get_run_manifest","name":"What one run was made from","description":"Reproducibility manifest for a run: code version and hash, sha256 of every input file, the labeled assumption set, engine, random seed / generator / draw count, output hash, and the ledger entry hash.","tags":["ledger","reproducibility","provenance"]},{"id":"run_payer_scenario","name":"Per-person acquisition-cost scenario (payer explorer)","description":"The site's payer explorer as a deterministic tool: published per-mL price benchmark (p25/p50/p75; US: CMS NADAC) x scenario daily dose x 365, with the oral side as a relative net price index. Inputs are clamped to the run's published ranges; regions without a staged price distribution return that absence, never a borrowed figure. covered_insulin_users is the caller's own multiplier, labeled as such. Scenario output, not a forecast.","tags":["scenario","payer","estimation"]},{"id":"ask_evidence_analyst","name":"Ask the Astra evidence analyst","description":"Natural-language answers over the published records, by GPT-6 Astra running a tool loop against the other skills here. The model supplies no figures of its own; each answer is labeled AI-generated and returns the tool-call trace it was built from, so every figure can be re-checked against its record. No efficacy, safety, regulatory, medical, or investment conclusion. Also at POST /ask for humans.","tags":["analyst","natural-language","gpt-6-astra"]}]}