Home · Developers · Contact · Privacy · Terms
CoAuth is an agent-native prior-authorization cockpit. It lets a clinician and an AI agent complete a health-insurance prior authorization together on one live page, using the WebMCP standard. Instead of an AI agent scraping the page and guessing where to click, CoAuth exposes the payer form as typed tools that the agent calls directly: filling fields, attaching evidence, scoring denial risk and detecting clinical conflicts.
The human stays in control, and that is enforced rather than asserted. The clinician signature is minted by the server over the exact submission being attested to, using a secret the browser never sees. The API then refuses any submission that is unsigned, that carries a forged token, or whose fields changed after signing. An agent running in the page cannot talk its way past that, because the check does not depend on anything the caller can set. This mirrors the legal reality that high-stakes clinical decisions require accountable human oversight.
Why it matters: in Medicare Advantage in 2024, 80.7% of appealed prior-authorization denials were fully or partially overturned, and only 11.5% of denials were appealed at all (KFF). Denials that are contested are usually overturned on appeal. The clinical decision is usually fine; the paperwork is not. CoAuth fixes the filing so patients get care sooner, while keeping a licensed human accountable for every submission. It was built for the OpenAI WebMCP Challenge as a demonstration of genuine human + agent collaboration on a real, high-stakes workflow.
The project is open source under the MIT license. See the source repository for the full implementation, including the 13 WebMCP tools, the heuristic denial-risk scoring and the conflict rules.