Cognizant ships 100+ MCP tools over TriZetto claims systems

Cognizant announced Workflow Agentic Processing and enterprise MCP servers for TriZetto Facets and QNXT today, shipping a library of more than 100 MCP tools over two of the oldest core administration platforms in US healthcare. The interesting part is not that a 30-year-old claims system now talks to agents. It is that the vendor is selling the audit trail as the feature.

What actually happened

Per Cognizant's announcement, Workflow Agentic Processing clears routine pended claims automatically and routes anything needing judgment to a human reviewer. Alongside it, Cognizant is shipping enterprise Model Context Protocol servers that expose more than 100 tools against TriZetto data context, so health plans can point their own AI work at customer service, prior authorization, and operations instead of building connectors.

The scale is the reason to pay attention: TriZetto platforms support over 200 million US healthcare members and process more than $500 billion in annual healthcare spend. Cognizant says the capability is available now for Facets and QNXT clients, with early-adopter plans piloting it.

Read the quote from Sanjay Subramanian, who heads the healthcare payer business: fewer pended claims, faster provider payments, and "full visibility into what the AI touched versus what the human reviewed." That last clause is a product requirement, not marketing. Prasad Sankaran, president of AI platforms, frames the same thing as scaling AI adoption "while maintaining governance and human oversight."

Why MCP servers on legacy systems matter for your business

You do not run Facets. You run QuickBooks, a Shopify admin, a PSA, a scheduling tool, an ERP that someone configured in 2014. Every one of those vendors is now on the same path: an MCP endpoint over the system of record, sold as an upgrade. Within a year you will be asked to turn one on.

When that lands, three questions decide whether it helps or hurts. First, what is the tool granularity? A hundred narrow tools you can scope is a better position than one run_query tool with write access, because narrow tools are the only thing that makes least privilege possible at the agent layer. Second, where does the work log live, and can you read it without the vendor? "What the AI touched versus what the human reviewed" has to be a queryable record on your side, or your dispute in March is a support ticket instead of evidence. Third, what happens to the exception path? The value here is not that routine claims clear — it is that the exceptions get routed to someone with authority. Automating the easy 70% and leaving the hard 30% in the same understaffed queue just moves the bottleneck and makes it less visible.

Copy the structure, not the stack: auto-handle the routine, route exceptions explicitly, log every action with an actor field that distinguishes agent from human. That is buildable on the tools you already pay for.

Sources: Cognizant newsroom, Cognizant TriZetto platform docs.