mcp

Find all access pathways for a GLP-1 drug (Ozempic, Wegovy, Mounjaro, Zepbound) given patient eligibility factors. Ranks pathways by monthly cost, flags prior-authorization rules and step-therapy, and returns the recommended appeal strategy if denial is likely. WHEN TO USE: User specifically asks about GLP-1 drugs (Ozempic, Wegovy, Mounjaro, Zepbound) — coverage, prior-auth strategy, step-therapy alternatives. WHEN NOT: For non-GLP-1 prescriptions (use optimize_prescription).

[Taxonomy VII.120 — baseline response (public data only)] Generate component arguments for an appeal letter: medical-necessity, contract, policy. WHEN TO USE: User has a denied claim and wants to appeal, or is preparing arguments / evidence / escalation steps for one. WHEN NOT: For explaining what a denial code means without drafting an appeal, use denial_code_explainer first. For finding a negotiation path on a non-denied bill, use negotiate_bill_script. EXAMPLES: - Argument blocks for CO-50 with UHC: `{"denial_code":"CO-50","insurer":"UnitedHealthcare"}`

[Taxonomy VII.129 — baseline response (public data only)] Find the correct escalation path for a denied claim (internal appeal → external review → DOI). WHEN TO USE: User needs to know the deadlines and agencies for each step of appealing a denial. Pass state (e.g. 'CA') and insurer for state-specific and insurer-specific values. WHEN NOT: Without state/insurer — still works but returns generic federal defaults. Follow up with state for best answer. EXAMPLES: - CA / UnitedHealthcare escalation path: `{"state":"CA","insurer":"UnitedHealthcare"}`

[Taxonomy E.25 — delegates to estimate_appeal_success] Given a denial, predict the probability of a successful appeal. Returns probability with 95% CI, sample size, and recommended appeal-strategy tags. WHEN TO USE: User or caller wants a probabilistic estimate (denial likelihood, appeal success, payment delay, etc.) to decide whether to pursue a path. WHEN NOT: For hard facts (dollar amounts, code definitions, deadlines). Predictions carry confidence bands; call out the uncertainty when you relay the number.

Scan multiple medical bills in one call. Returns per-bill anomaly lists + total recovery estimate across the batch. Input is a list of {bill_text, provider?, insurer?} items (max 10 per call). Much faster than serial scan_bill_for_errors when handling a bulk upload. WHEN TO USE: User has multiple bills or EOBs to review at once (e.g. a family's bills from one visit, or a year of statements). WHEN NOT: For a single bill (use scan_bill_for_errors).

[Taxonomy X.179 — baseline response (public data only)] Predict risk that a planned service will trigger an unexpected large bill. WHEN TO USE: User planning a service and wants to know if it will trigger an unexpected-large-bill situation. Pass expected_charge_usd and optionally cpt_code to get Medicare-ratio context. WHEN NOT: For a bill already received — scan_bill_for_errors first. EXAMPLES: - Check shock risk on $5,000 office visit: `{"expected_charge_usd":5000,"cpt_code":"99213"}`

[Taxonomy III.51 — baseline response (public data only)] Delta between billed charge and payer-allowed amount for a claim line. WHEN TO USE: User wants to understand what a service typically costs, compare prices across regions/providers, or gauge whether a billed amount is reasonable. WHEN NOT: For a specific bill they have in hand, scan_bill_for_errors is more actionable. For insurance-path estimates, use deductible_impact_calculator or coinsurance_simulator.

[Taxonomy C.16 — delegates to scan_bill_for_errors] Detect CPT unbundling (billing components of a procedure separately to inflate the claim). Uses the CMS NCCI edits plus payer-specific bundling rules. WHEN TO USE: Caller is building or validating a claim payload before submission — format, structure, completeness, alignment. WHEN NOT: For an already-submitted and denied claim, use generate_appeal_letter. For market pricing context, use benchmark_payer_rate.

[Taxonomy III.39 — baseline response (public data only)] Estimate a fair cash price for a service in a given region (no-insurance path). WHEN TO USE: Uninsured/self-pay user wants a fair-market cash price for a service. Returns Medicare national as a floor + commercial-multiple band. WHEN NOT: For insurance-covered scenarios (deductible_impact_calculator + coinsurance_simulator). EXAMPLES: - Cash price for office visit: `{"cpt_code":"99213","zip3":"902"}`

[Taxonomy H.41 — baseline response (public data only)] For a given service + patient plan, determine whether cash-pay + charity-care beats insurance billing. WHEN TO USE: User is planning for a medical expense, comparing plan options, looking for financial assistance, or trying to structure a payment plan. WHEN NOT: For billing-error analysis (scan_bill_for_errors) or appeal drafting (generate_appeal_letter).

[Taxonomy C.14 — delegates to scan_bill_for_errors] Detect internal contradictions in a claim: time mismatches, mutually exclusive codes, missing required fields, patient-gender/procedure mismatches. WHEN TO USE: Caller is building or validating a claim payload before submission — format, structure, completeness, alignment. WHEN NOT: For an already-submitted and denied claim, use generate_appeal_letter. For market pricing context, use benchmark_payer_rate.

[Taxonomy IV.75 — baseline response (public data only)] Detect duplicated service lines within a single claim. WHEN TO USE: Caller is building or validating a claim payload before submission — format, structure, completeness, alignment. WHEN NOT: For an already-submitted and denied claim, use generate_appeal_letter. For market pricing context, use benchmark_payer_rate.

[Taxonomy VII.128 — baseline response (public data only)] Generate a rebuttal to a payer's denial rationale point-by-point. WHEN TO USE: User has a denied claim and wants to appeal, or is preparing arguments / evidence / escalation steps for one. WHEN NOT: For explaining what a denial code means without drafting an appeal, use denial_code_explainer first. For finding a negotiation path on a non-denied bill, use negotiate_bill_script.

[Taxonomy VII.132 — baseline response (public data only)] Build a claim reconsideration request (lighter-weight than a formal appeal). WHEN TO USE: User has a denied claim and wants to appeal, or is preparing arguments / evidence / escalation steps for one. WHEN NOT: For explaining what a denial code means without drafting an appeal, use denial_code_explainer first. For finding a negotiation path on a non-denied bill, use negotiate_bill_script.

[Taxonomy II.25 — baseline response (public data only)] Check a set of CPT codes against NCCI bundling rules (distinct from modifier-based unbundling). WHEN TO USE: User asks about medical codes: what a CPT/ICD/HCPCS code means, whether a code combination is valid, or how to map between them. WHEN NOT: For full bill scans (use scan_bill_for_errors). For appealing a coding-related denial (use generate_appeal_letter).

[Taxonomy A.5 — delegates to scan_bill_for_errors] Detect CPT unbundling, mutually exclusive code pairs, and NCCI edit violations in a set of codes. Implemented by the bill scanner's anomaly engine. WHEN TO USE: User asks about medical codes: what a CPT/ICD/HCPCS code means, whether a code combination is valid, or how to map between them. WHEN NOT: For full bill scans (use scan_bill_for_errors). For appealing a coding-related denial (use generate_appeal_letter).

[Taxonomy III.49 — baseline response (public data only)] Simulate member coinsurance exposure across plan-design variants. WHEN TO USE: User wants to model their coinsurance share of a post-deductible charge. Pass charge_usd and optionally coinsurance_rate. WHEN NOT: For flat-copay services (copay_model). EXAMPLES: - Simulate 20% coinsurance on $1K: `{"charge_usd":1000,"coinsurance_rate":0.2}`

[Taxonomy III.50 — baseline response (public data only)] Model member copay for a service by plan-design class. WHEN TO USE: User wants the typical copay for a service type (primary / specialist / er). Uses KFF EHBS 2024 averages. WHEN NOT: For patient-specific plan details — copay varies by plan design. EXAMPLES: - Typical specialist copay: `{"service_type":"specialist"}`

[Taxonomy II.21 — baseline response (public data only)] For a given CPT, surface the ICD-10 codes typically paired with it for medical-necessity support. WHEN TO USE: User has a CPT code and wants to know what ICD-10 diagnoses commonly pair with it for medical-necessity support. WHEN NOT: For specific payer policy lookups (code_validation). EXAMPLES: - Common ICDs for 99213: `{"cpt_code":"99213"}`

Decode a CARC (Claim Adjustment Reason Code) denial code into plain English. Returns meaning, category, whether it's typically reversible on appeal, and appeal guidance. Source: public X12 External Code List + CMS/ERISA public rules. Free, no auth. WHEN TO USE: Reference lookups: individual code details, verifying a provider registry entry, checking a prior Medigami response's signature, or tracking an outcome. WHEN NOT: For composite workflows use the higher-level tool that calls these internally.

[Taxonomy III.48 — baseline response (public data only)] Estimate how a planned service will draw against the member's deductible. WHEN TO USE: User wants to know how a specific charge affects their deductible (how much draws down, remaining after, post-deductible exposure). WHEN NOT: For total annual out-of-pocket across a year (healthcare_budget_forecaster). EXAMPLES: - Compute deductible drawdown for $4K charge: `{"charge_usd":4000,"deductible_remaining_usd":1735}`

[Taxonomy VII.131 — baseline response (public data only)] Explain a CARC/RARC denial code in plain language with common causes. WHEN TO USE: User asks 'what does CO-16 mean' or similar — any CARC/RARC code. Returns full X12 table entry when available with meaning, category, typical root cause, primary remediation, and reversibility flag. WHEN NOT: For drafting an appeal (generate_appeal_letter). For the full escalation path (appeal_escalation_path_finder). EXAMPLES: - Explain CARC CO-16: `{"code":"CO-16"}` - Explain CARC CO-50 (medical necessity): `{"code":"CO-50"}`

[Taxonomy VI.106 — baseline response (public data only)] Classify a denial letter or remittance advice into normalized denial-reason categories. WHEN TO USE: User or caller wants a probabilistic estimate (denial likelihood, appeal success, payment delay, etc.) to decide whether to pursue a path. WHEN NOT: For hard facts (dollar amounts, code definitions, deadlines). Predictions carry confidence bands; call out the uncertainty when you relay the number.

[Taxonomy VII.121 — baseline response (public data only)] Build a denial-response package tailored to the denial reason code. WHEN TO USE: User has a denied claim and wants to appeal, or is preparing arguments / evidence / escalation steps for one. WHEN NOT: For explaining what a denial code means without drafting an appeal, use denial_code_explainer first. For finding a negotiation path on a non-denied bill, use negotiate_bill_script. EXAMPLES: - Build response package for CO-197: `{"denial_reason":"CO-197","insurer":"Aetna","cpt_code":"45378","amount":1850}`

[Taxonomy VII.123 — baseline response (public data only)] Assemble a complete dispute case file including claim, denial, evidence, and timeline. WHEN TO USE: User has a denied claim and wants to appeal, or is preparing arguments / evidence / escalation steps for one. WHEN NOT: For explaining what a denial code means without drafting an appeal, use denial_code_explainer first. For finding a negotiation path on a non-denied bill, use negotiate_bill_script.

Probability an appeal will be approved, derived from outcome-labeled data weighted by label provenance (IRO determinations, EOB reversals, physician confirmations). Returns probability + 95% CI + sample_size + expected_days_to_resolution + tracking_id. Report the actual outcome later via POST /api/outcomes/record with the same tracking_id to improve future estimates. WHEN TO USE: User or caller wants probability-of-reversal before spending time on an appeal. Returns probability + confidence interval. WHEN NOT: For picking among multiple appeal strategies (appeal_strategy_success_model). As a stand-alone — usually chain into generate_appeal_letter. EXAMPLES: - Gauge CO-50 appeal odds on Aetna: `{"insurer":"Aetna","denial_reason":"CO-50","cpt_code":"72148","state":"TX","amount":1200}`

[Taxonomy VII.124 — baseline response (public data only)] Generate an evidence package supporting a claim or appeal (references, studies, policies). WHEN TO USE: User has a denied claim and wants to appeal, or is preparing arguments / evidence / escalation steps for one. WHEN NOT: For explaining what a denial code means without drafting an appeal, use denial_code_explainer first. For finding a negotiation path on a non-denied bill, use negotiate_bill_script.

Human-readable rationale for an appeal probability. Returns {probability, tier, recommended_posture, rationale, expected_days_to_resolution, tracking_id} grounded in aggregate outcome rates — not model feature weights. Use this when the caller wants 'why' in plain English alongside the number. WHEN TO USE: Reference lookups: individual code details, verifying a provider registry entry, checking a prior Medigami response's signature, or tracking an outcome. WHEN NOT: For composite workflows use the higher-level tool that calls these internally.

[Taxonomy B.8 — delegates to benchmark_payer_rate] Fair-market price distribution for a service: DP-noised percentiles (p10, p25, p50, p75, p90) for the given CPT in a ZIP3 region. WHEN TO USE: User wants to understand what a service typically costs, compare prices across regions/providers, or gauge whether a billed amount is reasonable. WHEN NOT: For a specific bill they have in hand, scan_bill_for_errors is more actionable. For insurance-path estimates, use deductible_impact_calculator or coinsurance_simulator.

[Taxonomy X.181 — baseline response (public data only)] Find applicable financial-assistance programs (hospital charity, 340B, state, pharma). WHEN TO USE: User needs help paying a medical bill. Pass household_income_usd and family_size for FPL-based eligibility flags across 6 programs (501(r) charity, Medicaid, PE, manufacturer PAPs, 340B, state pools). WHEN NOT: For insured users with low exposure (use coinsurance_simulator or deductible_impact_calculator). EXAMPLES: - Find assistance for family of 3 at $45K income: `{"household_income_usd":45000,"family_size":3}`

[Taxonomy X.176 — delegates to financial_risk_exposure_model] Model a member's financial exposure. Alias of financial_risk_exposure_model. WHEN TO USE: User is planning for a medical expense, comparing plan options, looking for financial assistance, or trying to structure a payment plan. WHEN NOT: For billing-error analysis (scan_bill_for_errors) or appeal drafting (generate_appeal_letter).

[Taxonomy H.42 — baseline response (public data only)] Quantify a patient's financial exposure to medical-debt risk given current plan, health status, and historical claim volatility. WHEN TO USE: User is planning for a medical expense, comparing plan options, looking for financial assistance, or trying to structure a payment plan. WHEN NOT: For billing-error analysis (scan_bill_for_errors) or appeal drafting (generate_appeal_letter).

Format an LLM-citable reference to a previously-attested Medigami MCP response. Input: either the full signed envelope that an earlier attested tool returned, OR just a tracking_id. Returns a short citation string, a public verification URL a third party can open to inspect and cryptographically verify the payload, and a long-form attribution block. Use this after any substantive Medigami tool call (scan, estimate, resolve, benchmark) so the user has a verifiable reference to the specific answer and its expiration date. WHEN TO USE: After a substantive tool call, generate a citable string + short URL the LLM can paste into its user-facing response for later verification. WHEN NOT: For non-substantive lookups where citation is unnecessary. EXAMPLES: - Generate a citable URL from a tracking_id: `{"tracking_id":"evt_abc123"}`

DEPRECATED ALIAS of format_citation — kept for backward compatibility with v0.1.7 clients. New code should call format_citation directly. WHEN TO USE: Reference lookups: individual code details, verifying a provider registry entry, checking a prior Medigami response's signature, or tracking an outcome. WHEN NOT: For composite workflows use the higher-level tool that calls these internally.

Generate a legally-grounded insurance appeal letter for a denied claim. Returns the letter as a template with [PATIENT NAME] and [PROVIDER NAME] placeholders (structured-only PHI posture). Includes legal citations, recommended attachments, and appeal-deadline calendar. WHEN TO USE: User has a specific claim denied and wants a drafted appeal letter. Requires insurer, denial_reason (CARC code), cpt_code, diagnosis (ICD-10), and dollar amount. Returns an attested letter with reversal rate, insurer-specific tips, recommended attachments. WHEN NOT: If the user only wants to understand a denial code, use denial_code_explainer first. If they haven't received a formal denial yet, don't draft — clarify status first. For non-CARC disputes (e.g. billing errors pre-adjudication), use scan_bill_for_errors. EXAMPLES: - Draft CO-16 appeal to UHC for office visit: `{"insurer":"UnitedHealthcare","d...

[Taxonomy F.28 — delegates to generate_appeal_letter] [Taxonomy alias] Generate a state-specific insurance appeal letter with UPL-safe disclaimers for 19 US states. Same impl as generate_appeal_letter. WHEN TO USE: User has a denied claim and wants to appeal, or is preparing arguments / evidence / escalation steps for one. WHEN NOT: For explaining what a denial code means without drafting an appeal, use denial_code_explainer first. For finding a negotiation path on a non-denied bill, use negotiate_bill_script.

External-review filing deadline for a state + the date of the FINAL adverse determination. Returns the deadline date, the statutory window label + citation, and whether the exact state-specific window still needs verification (most states carry a conservative NAIC-model 4-month floor unless statute-confirmed, e.g. CA/NY). Free, no auth. WHEN TO USE: Reference lookups: individual code details, verifying a provider registry entry, checking a prior Medigami response's signature, or tracking an outcome. WHEN NOT: For composite workflows use the higher-level tool that calls these internally.

Where and how to file an external review for a given state — the venue name, filing channel (portal/mail/fax), portal URL, and whether expedited review is available. Free, no auth. WHEN TO USE: Reference lookups: individual code details, verifying a provider registry entry, checking a prior Medigami response's signature, or tracking an outcome. WHEN NOT: For composite workflows use the higher-level tool that calls these internally.

[Taxonomy X.183 — baseline response (public data only)] Forecast a household's annual healthcare spend based on plan + utilization. WHEN TO USE: User planning annual healthcare spend, comparing plans, or estimating worst-case exposure. WHEN NOT: For a specific procedure cost (benchmark_payer_rate). EXAMPLES: - Forecast annual spend range: `{}`

[Taxonomy II.22 — baseline response (public data only)] For a given ICD-10, surface the CPT codes typically billed for that diagnosis. WHEN TO USE: User has an ICD-10 diagnosis and wants to know what CPTs are commonly billed for it. WHEN NOT: For planning specific services (use benchmark_payer_rate once the CPT is chosen). EXAMPLES: - Common CPTs for essential hypertension: `{"icd10_code":"I10"}`

[Taxonomy III.46 — baseline response (public data only)] Inflation-adjusted price comparison across years for a CPT. WHEN TO USE: User has a historical dollar amount and wants it adjusted for medical-care inflation to a given year. WHEN NOT: For current market prices (benchmark_payer_rate). EXAMPLES: - Adjust $1K 3yr forward at medical CPI: `{"base_amount_usd":1000,"years":3}`

[Taxonomy H.39 — baseline response (public data only)] Given a patient's historical claim patterns, rank available insurance plans by expected annual cost. WHEN TO USE: User is planning for a medical expense, comparing plan options, looking for financial assistance, or trying to structure a payment plan. WHEN NOT: For billing-error analysis (scan_bill_for_errors) or appeal drafting (generate_appeal_letter).

[Taxonomy F.31 — baseline response (public data only)] Parse an insurer's response letter or EOB into structured fields: denial reason codes, appeal rights, deadlines, next-step recommendations. WHEN TO USE: User has a denied claim and wants to appeal, or is preparing arguments / evidence / escalation steps for one. WHEN NOT: For explaining what a denial code means without drafting an appeal, use denial_code_explainer first. For finding a negotiation path on a non-denied bill, use negotiate_bill_script.

[Taxonomy X.185 — baseline response (public data only)] Recommend insurance-plan selections to maximize value given projected utilization. WHEN TO USE: User is planning for a medical expense, comparing plan options, looking for financial assistance, or trying to structure a payment plan. WHEN NOT: For billing-error analysis (scan_bill_for_errors) or appeal drafting (generate_appeal_letter).

Look up a CPT or HCPCS-J code or search by keyword. Covers the most common outpatient billing codes (E/M, preventive, lab, imaging, procedure, cardiology, therapy, behavioral, vaccine, injectable drugs). Note: CPT is a registered trademark of the American Medical Association; full-dataset commercial use requires an AMA license. This tool provides short descriptors for identification in billing-review workflows. WHEN TO USE: User asks about a specific CPT/HCPCS procedure code or wants to search for codes by procedure description. WHEN NOT: For the expected ICD pairings (icd_to_cpt_mapper or cpt_to_icd_mapper). EXAMPLES: - Find CPT for routine office visit: `{"query":"office visit established patient"}`

Look up an ICD-10-CM diagnosis code or search by keyword. Input is either an exact code (e.g. 'E11.21') or a free-text description phrase ('type 2 diabetes with nephropathy'). Returns either the matching code + description or up to 20 candidate matches. Use when a clinical note uses non-standard phrasing and the correct code is needed for billing. WHEN TO USE: User asks about a specific ICD-10 diagnosis code or wants RAG-search over ICD-10 for a condition name. WHEN NOT: For the matching CPT codes (cpt_to_icd_mapper). For code validation against a payer (code_validation). EXAMPLES: - Find ICD-10 for hypertension: `{"query":"essential hypertension"}`

Look up a provider in the CMS NPPES National Provider Identifier registry. Accepts a 10-digit NPI number OR a provider name (optionally narrowed by 2-char state). Returns up to 20 matches with entity type, specialty, taxonomy code, address, and status. Source: CMS NPPES public API (free, no key required). WHEN TO USE: User wants to verify a specific provider by NPI number, or find a provider by name + state. WHEN NOT: For provider performance analytics (provider_efficiency_score). EXAMPLES: - Look up a provider by NPI number: `{"npi_number":"1234567893"}`

Look up a NUCC provider taxonomy code (the specialty codes used on HIPAA transactions + NPI registrations) or search by keyword. Returns classification + type + optional specialization. Source: NUCC Health Care Provider Taxonomy Code Set (public; CMS-accepted). WHEN TO USE: Resolving a NUCC provider taxonomy code to its specialty description. WHEN NOT: For finding a provider by specialty in a region (not yet implemented).

[Taxonomy III.55 — baseline response (public data only)] Full DP-noised market price distribution for a CPT+region pair. WHEN TO USE: User wants to understand what a service typically costs, compare prices across regions/providers, or gauge whether a billed amount is reasonable. WHEN NOT: For a specific bill they have in hand, scan_bill_for_errors is more actionable. For insurance-path estimates, use deductible_impact_calculator or coinsurance_simulator.

Given a prior scan_id, returns a prioritized action list: anomalies ranked by (expected overcharge x recovery probability), with the next steps to pursue each. Use after scan_bill_for_errors. WHEN TO USE: User has already run scan_bill_for_errors and wants the anomalies ranked by expected dollar recovery. WHEN NOT: As a first call — always scan first.