Agent skills hit one million registrations while the business context problem gets a production-grade answer. This week shows the infrastructure layer maturing fast: standardized skill distribution at scale and governed access to trusted enterprise data through the Model Context Protocol.
The Genie One MCP is now Generally Available
Databricks moved Genie One MCP into general availability with Unity Gateway managing access, audit logs, and fine-grained policies for every agent connection. This solves the critical problem of agents returning conflicting answers because they lack shared business definitions—your Claude instance and your team's ChatGPT workspace now query the same governed ontology instead of interpreting raw tables independently. The MCP server acts as a single integration point, so you define metrics, relationships, and access rules once in Databricks and enforce them across every tool your organization uses.
// MCP client connecting to governed Genie ontology
const client = new MCPClient({
server: "genie-one-mcp.databricks.com",
credentials: process.env.UNITY_GATEWAY_TOKEN
});
// Agent queries "revenue by region" — returns governed answer
const response = await client.queryMetric("revenue_by_region", {
timeRange: "Q3_2026",
filters: { region: ["EMEA", "APAC"] }
});State of agent skills
Skills.sh reached one million registered agent skills in seven months with 280 million installs, crushing early growth rates for GitHub, npm, and mobile app stores. The distribution is extreme: 0.04% of skills account for 62% of installs, and cross-industry capabilities like "parse structured logs" or "generate compliance reports" get installed 10x more than vertical-specific tools. The data shows that reusable, portable skills win—developers want general-purpose building blocks, not narrow industry solutions that lock them into specific workflows or domains.
Genie One MCP: Give any AI Agent the Right Business Context
This post breaks down how Genie Ontology provides semantic understanding on top of raw data access, turning fragmented tables into a unified business context layer that any MCP-compatible agent can query. Without this layer, agents either hallucinate metric definitions or return technically accurate but business-meaningless results because they lack knowledge of approved relationships, calculation rules, and access boundaries. Databricks positions this as "define once, operationalize everywhere"—you maintain a single source of truth for business logic and push it to every agent tool across your org, avoiding the consistency nightmare of per-tool configuration or unrestricted data access.
The skills registry numbers confirm what production teams already know: agent value scales through reusable capabilities, not bespoke tools. Meanwhile, MCP-based governance infrastructure like Genie One solves the trust and consistency problems that blocked enterprise agent adoption six months ago. 🛠️