The integration barrier that kept AI underwriting at the top of the market is gone. Here’s what we’ve done about it.
For two years, the story of AI in commercial insurance belonged to the largest carriers — the ones who could hire the engineers, build the infrastructure, and fund the multi-quarter integration projects. Everyone else asked the same question: how do we get there without building what they built?
They don’t have to anymore.
The barrier was never the AI — the models were ready. It was the plumbing: the custom integration layer between a vendor’s product and a carrier’s stack. Every stack was different, so every deployment was bespoke, slow, and expensive. That’s what kept AI-augmented underwriting concentrated at the top.
A protocol changed the math
The Model Context Protocol (MCP) is an open standard introduced by Anthropic in November 2024 for connecting AI assistants to the systems where data lives. In under a year it became infrastructure. Major AI providers including OpenAI and Google DeepMind adopted it; Microsoft built it into Copilot and Windows; the community shipped thousands of MCP servers and the industry settled on it as the de-facto standard for connecting agents to tools and data. In December 2025 Anthropic handed MCP to a Linux Foundation body whose founding members include OpenAI, Google, Microsoft, AWS, Bloomberg, and Cloudflare.
The effect is simple: a carrier already running Copilot, or an agent built on Claude or OpenAI, can now call an MCP-enabled service straight from those tools. No middleware. No proprietary connector. No quarter-long project. The AI investment already made becomes the way to deliver capabilities never built.
What we’ve brought to that moment
NeuralMetrics now speaks MCP through a single hosted server — putting authoritative commercial risk and classification data one connection away from any MCP-compatible client: Claude Desktop, Web, and Mobile, Claude Code, Cursor, and more.
An underwriter’s agent can submit a company, classify it, pull risk insights, and check appetite and eligibility — in one workflow, grounded in NeuralMetrics data, not the open internet. Submit by name and address — or by website for Risk Intelligence — poll for results, retrieve full company details and risk decisions, all returned as structured, schema-validated responses an agent can act on directly.
And because this data drives real underwriting decisions, it’s built to production standards: OAuth 2.1 access brokered through our own authentication, no credentials stored or exposed at the edge, entitlement-aware so agents see only what a user is licensed for, and every response redacted and error-classified so an agent knows a bad address from an entitlement limit from a retryable blip.
NeuralMetrics is live on MCP now — connect Claude Desktop, Claude Code, Cursor, or any MCP client directly, no integration project required. Full production rollout completes this week, with a developer binary already available for teams that want to self-host, and listing in the Anthropic Connectors Directory and the official MCP Registry coming soon.
Explainability Matters
Of course, automation alone isn’t enough. Trust matters most.
In regulated industries like insurance, every decision must be explainable—to regulators, to customers, and to the teams making them. Achievable AI meets that standard by showing its work.
When the system classifies a business, it reveals the reasoning behind it: the signals considered, the context applied, and the risk attributes detected. There are no black boxes—just clarity.
That transparency does more than satisfy compliance. It creates alignment.
When underwriters, agents, and customers can see how a conclusion was reached, trust follows naturally. It also reduces the time and friction spent reconciling discrepancies later in the process, because everyone starts from the same source of truth.
Explainability isn’t a technical feature; it’s an ethical one. It ensures that AI supports human expertise instead of obscuring it. And for carriers of any size, that’s the kind of foundation worth building on.
Why this reshapes the market
Large carriers gain flexibility — they can compose vendors more freely and cut lock-in. But that only optimizes a program they already have.
For everyone else — leaner teams, smaller IT, tighter budgets — this removes the one barrier that mattered. No integration layer to build. No AI team to hire. No systems-integration engagement to fund. A build project becomes a deployment.
That redraws the competitive line. The edge no longer belongs to whoever built the most sophisticated AI stack. It belongs to whoever puts the right capability at the right decision, inside the workflows their team already uses. The narrative around AI in insurance has been about what it can do. The next chapter is about who gets to use it — and we’ve just widened that circle considerably.
This is a foundation, not a finish line — with more coming, including guided workflow prompts for common underwriting journeys. For carriers who waited because the barrier felt too high, the wait is over. The question is no longer whether to adopt. It’s what to adopt first, and where to put it.
Achievable AI isn’t the future—it’s what’s possible right now.