MCP servers for AI agents
9 production pre-action gates for orchestral agents. Each server intercepts your agent at a decision point before a consequential action — payment execution, supplier onboarding, data transmission, customs filing, document acceptance. Returns a machine-readable verdict with agent_action field. No human checkpoint required. Built for agentic finance workflows including Robinhood Agentic Trading and autonomous payment pipelines.
Pre-action gate pattern
Each server follows the same pattern. Your orchestrator reaches a decision point. It calls the relevant gate. The gate returns a structured verdict with agent_action field. The orchestrator routes based on verdict — no LLM reasoning required at the gate layer.
Bizfile MCP — Counterparty Validator
Pre-condition: agent has identified a new counterparty and has not yet executed payment or contract. Call validate_counterparty for KYC confidence HIGH/MEDIUM/LOW, AI risk score 0-100, company officers, and PROCEED/EDD/BLOCK verdict. Call screen_counterparty with the officers array to check all directors against 386 risk data sources including sanctions, PEPs, and watchlists. A missed sanctions match on an autonomous payment creates criminal liability with no recovery path.
| tool | tier | what it returns |
|---|---|---|
| validate_counterparty | free · 20/month | Registry lookup, active status, KYC confidence, AI risk score 0-100, officers list, PROCEED/EDD/BLOCK verdict |
| validate_counterparty_lite | free · 20/month | Fast registry-only check, no AI scoring. Company status, registration number, address. Sub-second response. Use for high-volume pre-screening. |
| screen_counterparty | paid | Screens company + all officers against 386 risk data sources (sanctions, PEPs, watchlists). PROCEED/HOLD verdict per entity. Criminal liability if skipped on autonomous payment. |
VAT Validator MCP
Pre-condition: agent has received a supplier invoice and has not yet approved payment. Call validate_vat with the VAT number and invoice company name — returns PROCEED/VERIFY_MANUALLY/HOLD verdict with fraud risk score 0-100, name match check, and country_code. Call get_vat_rates with country_code to confirm the correct rate. Approving payment against an invalid VAT number creates unrecoverable tax liability.
| tool | tier | what it returns |
|---|---|---|
| validate_vat | free · 50/month | EU VIES, UK HMRC, AU ABR validation. Fraud risk score 0-100. Name cross-check. PROCEED/VERIFY_MANUALLY/HOLD verdict. |
| get_vat_rates | free · 50/month | Current standard and reduced VAT rates for all 27 EU member states, UK, Australia. Call after validate_vat with country_code. |
Tender MCP
Pre-condition: agent needs to identify active procurement opportunities or evaluate bid viability. Call search_tenders for live contracts across UK Contracts Finder, EU TED, and US SAM.gov with AI fit score and BID/INVESTIGATE/SKIP verdict per result. Call get_tender_intelligence for DAILY_DIGEST monitoring or AWARD_HISTORY research.
| tool | tier | what it returns |
|---|---|---|
| search_tenders | free · 10/month | Live contracts from UK/EU/US. AI fit score 0-100 when company_profile provided. BID/INVESTIGATE/SKIP verdict per result. |
| get_tender_intelligence | paid | DAILY_DIGEST: new tenders last 24h. AWARD_HISTORY: past contract winners. Machine-readable agent_action field. |
URL Safety Validator MCP
Pre-condition: agent has a URL from an untrusted source and is about to fetch, visit, or forward it. Call check_url before every URL from emails, scraped pages, user inputs, documents, or API responses. A DANGEROUS verdict means stop immediately — one weaponised URL followed by an agent equals credential exfiltration with no recovery path.
| tool | tier | what it returns |
|---|---|---|
| check_url | free · 10/month | SAFE/SUSPICIOUS/DANGEROUS verdict, AI trust score 0-100, threat categories, SSL status, domain age, redirect chain flag. Cross-checks Google Web Risk + Google Safe Browsing + AI analysis. |
Data Compliance Classifier MCP
Pre-condition: agent has assembled a data payload and is about to pass it to an external API, database, or third-party service. Call validate_data_safety before any external transmission. Returns SAFE_TO_PROCESS, REDACT_BEFORE_PASSING, DO_NOT_STORE, or ESCALATE verdict with applicable regulations. Call get_safety_report after a non-SAFE verdict for redaction strategy and compliance-safe reformulation.
| tool | tier | what it returns |
|---|---|---|
| validate_data_safety | free · 20/month | PII classification, GDPR/HIPAA/PCI-DSS/CCPA detection, breach check, compliance verdict. SAFE_TO_PROCESS/REDACT_BEFORE_PASSING/DO_NOT_STORE/ESCALATE. |
| validate_data_safety_lite | free · 20/month | Pattern detection only. No AI call. Sub-100ms. SAFE_TO_PROCESS/REVIEW_REQUIRED. Use for high-volume pre-screening before full classification. |
| get_safety_report | paid | Redaction strategy, compliance-safe reformulation, batch up to 50 payloads, audit report. Call after non-SAFE verdict. |
Local Model Suitability MCP
Pre-condition: agent is about to route a task to a cloud model. Call check_local_viability first. If verdict is LOCAL, skip the cloud call entirely — saves cost on every routable inference call. Pass data_sensitivity=CONFIDENTIAL to force LOCAL verdict for tasks that must not leave the machine.
| tool | tier | what it returns |
|---|---|---|
| check_local_viability | free · 20/month | LOCAL/CLOUD/EITHER verdict, recommended Ollama model, estimated cost saving, confidence score |
HS Code Classifier MCP
Pre-condition: agent is about to generate an invoice, customs declaration, or duty calculation. Call hs_classify_product to get the verified 6-digit HS code before filing. Call hs_validate_code to verify any supplier-provided code before submitting to a customs authority. An incorrect HS code on a customs declaration creates a compliance violation regardless of intent.
| tool | tier | what it returns |
|---|---|---|
| hs_classify_product | free · 10/month | HS code, WCO 6-digit parent, official source, tariff version, AI classification reasoning, agent_action field |
| hs_validate_code | paid | Validates supplier-provided HS code — VALID/INVALID/MISMATCH/OUTDATED verdict, product match score, mismatch reason, correct code suggestion |
Quantum Suitability Validator MCP
Pre-condition: agent is evaluating whether to route a computational problem to a quantum provider or allocate quantum POC budget. Call quantum_assess_problem for four independent scores — scientific fit, hardware feasibility, advantage potential, commercial relevance — and an advantage_claim_level from NONE to PRODUCTION_VALIDATED. The scoring architecture was validated by a domain expert. SCIENTIFICALLY_RECOMMENDED_NOW means the problem is worth quantum investigation. COMMERCIALLY_RECOMMENDED_NOW means defensible commercial deployment today — a high bar. Call quantum_readiness_report with profile=RESEARCH, ENTERPRISE, or INVESTOR for hardware requirements and provider recommendations.
| tool | tier | what it returns |
|---|---|---|
| quantum_assess_problem | free · 5/month | Four independent scores. advantage_claim_level: NONE through PRODUCTION_VALIDATED. Verdict: SCIENTIFICALLY_RECOMMENDED_NOW / COMMERCIALLY_RECOMMENDED_NOW / INVESTIGATE_FURTHER / PREMATURE / NOT_QUANTUM_AMENABLE. |
| quantum_readiness_report | paid | Pass profile=RESEARCH, ENTERPRISE, or INVESTOR. Returns hardware requirements, qubit estimates, error correction needs, recommended providers weighted by decision context. |
Document Integrity Validator MCP
Pre-condition: agent has received a document and is about to act on its contents — approve a payment, onboard a counterparty, or submit to a regulatory authority. Call check_document before acting. A FAIL verdict means the document contains internal inconsistencies indicating tampering — do not proceed. Call check_document_package for multi-document trade finance workflows.
| tool | tier | what it returns |
|---|---|---|
| check_document | free · 10/month | Checks a single document against its international standard — PASS/FLAG/FAIL/UNKNOWN_DOCUMENT_TYPE verdict, flags, confidence score |
| check_document_package | paid | Checks 2-20 related documents and cross-checks for consistency conflicts — weights, party names, reference numbers, dates, commodity descriptions, port references |