Ketryx, a compliance platform for medical devices and defense contractors, has launched a beta of its Model Context Protocol server — and the angle isn't the protocol. It's what the protocol is connecting.
On March 31, 2026, the Cambridge, Massachusetts-based company announced the beta launch of an MCP server that brings its compliance intelligence graph directly into AI agent workflows NewsFile Corp. The MCP protocol, which Anthropic introduced in late 2024 and released as open source, lets AI tools query external data sources in a structured way. Ketryx's implementation connects ChatGPT, Claude, and Copilot to its knowledge graph of product lifecycle data — the tangled web of requirements, design specifications, test records, and failure reports that regulated industries generate in volumes regulators actually read.
The framing matters: Ketryx isn't selling AI that handles compliance. It's selling the plumbing that makes AI usable in compliance-adjacent work, where the documentation isn't overhead — it's the product.
Ketryx CEO Erez Kaminski put the paradox plainly in a company blog post: "Regulated companies are stuck in a paradox. The systems that are meant to ensure safety also slow teams down to the point where their best engineers are spending more time formatting documents than solving patient-critical problems." The company's MCP implementation is the infrastructure-level answer to that paradox — making the fragmented compliance graph queryable at inference time rather than in retrospect.
Ketryx raised a $39 million Series B in September 2025, bringing total funding to over $55 million Ketryx press release. The round was led by Transformation Capital, with participation from Bill Hawkins, former CEO of Medtronic. Hawkins' involvement is worth noting: he ran a company that spent decades accumulating exactly the kind of compliance debt Ketryx is pitching against. "I saw how much time our teams spent proving compliance versus creating value," he said in the release. "The opportunity to flip that ratio is significant."
Ketryx's background is medical devices specifically. The company's VP of Regulatory Affairs, Paul Jones, co-authored IEC 62304, the international standard governing medical device software lifecycle processes Ketryx blog. When Jones talks about validated agents, he's operating from a standard that predates generative AI by a decade and was written with human practitioners in mind. Ketryx's position is that IEC 62304 is a feature, not a constraint — that fitting AI into its architecture produces agents more likely to pass regulatory scrutiny than agents built without it.
The MCP integration addresses a real friction point in enterprise AI deployment. General-purpose AI agents lack access to the compliance context that makes regulated work tractable. A model asked to review a design specification for a Class III medical device has no built-in way to check that specification against the applicable requirements, the risk analysis, or the test protocol that validates it. MCP solves the connectivity problem. Ketryx's knowledge graph provides the structured data underneath.
Ketryx claims its AI agents cut manual compliance work by 90 percent NewsFile Corp. That's a Ketryx figure on Ketryx outcomes, and Giskard will want to see the methodology before it becomes a reference point in every pitch deck for the next eighteen months. But the directional claim is consistent with what regulated industry AI adoption looks like when it works: not replacing compliance officers, but eliminating the documentation overhead that consumes their time.
The "validated agent" framing is doing real work in Ketryx's positioning. The company distinguishes between general AI agents and agents that have been defined, demonstrated, and documented for specific regulatory tasks Ketryx blog. "When we say an agent is validated, we mean we have defined a specific task and we have demonstrated, with audit-ready evidence, that the agent can perform that task reliably and safely," the company explains. That language maps directly to IEC 62304 vocabulary — the standard distinguishes between software items (discrete units of functionality) and software systems (the integrated whole), and both require documented evidence of lifecycle conformance.
The MCP beta is a narrow technical release with a specific audience: teams at medical device and defense companies building with AI tools who need compliance-aware context at inference time. Whether that audience is large enough to support a $55M-plus company is an open question. But the architecture Ketryx is building — connecting structured compliance data to agentic AI via an open protocol standard — is a pattern that could apply across any heavily regulated industry. Pharmaceuticals, financial services, nuclear: all have the same graph problem, the same documentation burden, the same regulatory pressure to prove AI systems are doing what the paper says they're doing.
Ketryx's bet is that the compliance graph is the competitive moat, not the AI model. That's a plausible bet. Regulated industries don't switch AI models because the new one is slightly better at reasoning. They switch when the compliance documentation is already done. MCP makes the graph accessible. The graph makes the compliance work legible. That's the chain Ketryx is building, and it's more interesting than another AI launch announcement.
The open question is whether "validated agent" becomes a regulatory category or stays a marketing claim. IEC 62304 was written for human-written software. Whether AI-generated decisions under IEC 62304 supervision satisfy regulators depends on interpretations that haven't been tested at scale. Ketryx is building as if the interpretation will be favorable. The bet is reasonable. The evidence is still accumulating.
† Add footnote: "Source-reported; not independently verified."
†† Add footnote: "Source-reported; not independently verified." Or attribute to Anthropic directly if the attribution can be confirmed.
