MCP

The Model Context Protocol, MCP, is an open standard for connecting AI systems to external tools and data sources. Rather than building a one-off integration for every AI application that wants access to claim data, Polygrade exposes an MCP server that any MCP-compatible AI agent or tool can connect to directly.

What Polygrade exposes over MCP

Polygrade’s MCP server surfaces read access to the same claim, authorization, and payment records Polygrade already tracks, model and serial numbers, triage conversations, authorization decisions and their confidence levels, parts orders, and payment status, so an AI agent operating elsewhere can query a claim’s current state without a custom API integration.

What an MCP client can do is scoped by your configuration, not open by default. Read access to claim data is the starting point; any action beyond that is explicitly enabled per deployment.

How the MCP connection works

MCP defines a standard way for a client, an AI agent, a chat interface, an internal tool, to discover and call the capabilities a server exposes. Polygrade’s MCP server implements that standard directly, so any compliant client can connect without Polygrade-specific integration code.

A Forward Deployed Engineer scopes exactly which claim data and actions are exposed through your MCP server during deployment, matching what your team wants other AI tools to be able to reach.

What stays with Polygrade

Claim adjudication, authorization decisions, and payment execution stay inside Polygrade exactly as they do today. MCP is a way for other tools to read that record, not a way to change how it’s produced.

Data handling and security

Every MCP connection runs against a company-specific Polygrade instance, one client’s claim data is never accessible to another. Polygrade is SOC 2 compliant, with both Type I and Type II audits complete.

Every query made through Polygrade’s MCP server is logged, so any AI agent or tool reading claim data through it leaves the same audit trail as a person would.

Deploying MCP support

A dedicated Forward Deployed Engineer configures your MCP server’s scope, which records and actions are exposed, and to which clients, before it goes live.

Frequently asked questions about Polygrade and MCP

Does exposing an MCP server give any AI agent access to claim data?

No. Access is scoped per deployment; only the records and actions your configuration allows are reachable, and only by clients you’ve connected.

How is this different from Polygrade’s other integrations?

Polygrade’s other integrations connect to specific third-party systems. MCP is a standard interface into Polygrade itself, so other AI tools can read claim data without a custom integration.

Can MCP be used to write data into Polygrade, not just read it?

Read access to claim, authorization, and payment records is the default. Any write capability is explicitly scoped and enabled per deployment.