Humans keep control. Agents get reliable tools.
AidPath turns a confusing eligibility website into an agent-readable capability surface. People can enter only the facts they choose; agents can analyze, compare and prepare next steps through structured WebMCP tools instead of guessing through the UI.
Shared human + agent workbench
This prototype uses illustrative support programs so the interaction model can be evaluated without presenting legal or financial advice. The architecture is designed to ingest verified official rules in production.
Waiting for a case
Run the form or ask a WebMCP-aware agent to analyze a structured case.
Why WebMCP changes the experience
Without WebMCP, an agent must infer labels, scrape rendered text, click controls and hope the page has not changed. With WebMCP, the site publishes a stable capability contract while the human still sees every meaningful result.
Discover
The agent discovers typed tools such as program listing, case analysis and comparison.
Execute
The tool calls the same deterministic engine as the visible UI and logs the action on screen.
Verify
The human sees matched and missing conditions and can correct facts before taking any real-world action.
Tool surface
The implementation uses the WebMCP Imperative API with JSON Schema contracts and explicit read-only annotations where appropriate.
Discover the available illustrative programs, optionally filtered by category.
Run intent classification and deterministic matching, then update the visible result panel.
Return matched and missing conditions for one option against a supplied profile.
Rank a selected set of options without relying on UI scraping.
Stage a sample case in the human interface so a person can inspect or edit it.
Produce a concise, visible checklist based on missing conditions for a selected option.