Agent-ready policy review with MCP, CLI, and OpenAPI

Agent Workflow

PolDex exposes schema discovery, classification, estimates, extraction, polling, artifacts, parser-output import, Reducto import, and benchmark status through agent-readable interfaces.

This is an infrastructure and interface case study. It does not claim autonomous production usage by a named customer.

Status	Agent interface case
Audience	AI agents, insurtech builders, broker workflow tools, and technical operators
Schema or surface	`agent`

All Case Studies Agent Page ->

Before / After

The job PolDex takes off the workflow.

The product boundary is narrow: turn messy insurance evidence or parser output into reliable, schema-controlled data that another system can use.

Before

Manual or generic extraction

Agents need machine-readable surfaces, not a human-only dashboard.
Uncontrolled agent extraction can spend credits without confirmation or call unsupported schemas.
Generic document outputs require extra glue before they can be trusted inside insurance workflows.

After

PolDex output

Agents can discover schema maturity, estimate costs, ask for confirmation, run extraction, poll jobs, and download artifacts.
MCP, CLI, OpenAPI, llms.txt, and the agent manifest all point to the same production contract.
FastScript keeps the output schema-scoped and evidence-backed for downstream tool use.

Workflow

How the proof path runs.

Discover

The agent reads llms.txt, OpenAPI, the agent manifest, hosted MCP metadata, and schema endpoints.

Estimate and confirm

The agent estimates credits and requires explicit confirmation before spending.

Extract and poll

The agent creates a job or batch, polls status, and handles webhook or connector events.

Download and route

The agent downloads JSON, CSV, XLSX, or ZIP artifacts and passes evidence-backed data to the next tool.

Output

What downstream systems can use.

These are the fields or artifacts this case study is meant to make legible to databases, workflows, operators, and AI agents.

Structured output

Data contract

schema_discovery
classification
credit_estimate
job_id
batch_status
artifact_links
benchmark_status
connector_events
parser_output_import
mcp_tool_response

Proof

Evidence contract

OpenAPI route contract
Agent manifest and llms.txt freshness
MCP tool metadata
Schema maturity and benchmark state
Credit confirmation fields

Try It