Tools (4)
| Tool | Access | What it does | Serves |
|---|---|---|---|
classify_document | Reads & writes | Detect boundaries, classify pages/segments and index document manifests. | 3 agent tools |
extract_fields | Read | Extract typed fields with confidences and provenance from a classified document. | 5 agent tools |
extract_financial_statements | Read | Extract financial line items with captions, periods and bounding boxes for spreading. | 1 agent tool |
store_evidence | Writes | Store claimant- or customer-supplied files and return safe metadata. | 1 agent tool |
Overview
One document-intelligence surface for every agent that reads documents: paystubs, deal jackets, financial statements, compliance certificates, ACORD submissions, KYC IDs and claim evidence. Returns classifications, extracted fields with confidences and page/bounding-box provenance so downstream engines and humans can trace every figure.
Backing systems
- Intelligent document processing (IDP) platform
- Document/evidence store
Security & data
Documents carry PII and NPI. Content is processed in-region and never retained by the MCP beyond the request; storage goes to the lender's document store. Field-level confidences are returned so agents route low-confidence items to humans.
Build & deploy
Wrap the IDP vendor's REST API (OpenAPI available for most vendors) and normalise output to a shared field/provenance schema. Good first candidate for the API-to-MCP builder.
Packaged as a container alongside the agents that use it; deployed into the licensee's environment.