Platform teams face a growing challenge as coding agents like Claude Code, Cursor, and Codex proliferate across engineering orgs: individual developers build powerful custom configurations (skills, MCP integrations) that never get shared, leading to inconsistent output quality. The post argues this is a knowledge distribution problem, not a model limitation. It presents Portkey as a centralized control plane for managing agent skills and MCP server configurations at scale — offering a Skills Registry with versioning and approval workflows, an MCP Registry with centralized auth, and a CLI that provisions everything for developers in one command. The Docker/Kubernetes analogy is used: coding agents are Docker, Portkey is the orchestration layer needed to run them at organizational scale.
Table of contents
The engineers who get more out of agentsWhat breaks as teams scaleUber's Platform team has already solved thisA unified registry for Skills and MCPThe platform team's roleGet startedSort: