Oversight Lints the Storybook Manifest Your AI Agent Actually Reads
Oversight, a new Storybook addon, lints the MCP manifest AI agents actually read, catching missing docs, extractor drift, and broken redirects.
A developer found that a coding agent, instead of reusing an existing design system, built two components from scratch. The cause: Storybook's MCP integration only reads the auto-generated components.json manifest, never the Docs tab, source comments, or JSDoc tags — so any guidance living outside that manifest is effectively invisible to the agent.
To catch these gaps, the author built Oversight, an open-source Storybook addon that lints the manifest for failure modes that silently strip context from agents: failed docgen extraction, mismatched docgen extractors (react-docgen versus react-docgen-typescript resolve TypeScript types differently), undocumented required props, missing component descriptions, and dangling redirect links.
Notably, component-level JSDoc tags like @deprecated never reach the agent's answer — only tag-stripped description prose does — so selection guidance has to live as plain sentences with real links rather than structured metadata. Oversight introduces its own @oversightIgnore tag, deliberately kept separate from Storybook's manifest-hiding !manifest tag, so linter exemptions don't accidentally remove a component from the agent's view.
For teams increasingly wiring AI agents into frontend workflows via MCP, this points to a broader gap: documentation pipelines built for humans don't automatically carry context to machine consumers, and nothing in a standard Storybook build currently flags that silently.