Notion-to-Claude Bridge

Notion is where many people store their accumulated knowledge. Claude is where they want to reason about it. The gap between the two is a context window — and context windows are finite. Dumping your entire workspace produces garbage: too much noise, critical details diluted, token budget wasted on formatting and metadata that doesn't help. The bridge has to be selective, structured, and aware of what Claude actually needs to be useful. This is a relevance engineering problem, not a data pipeline one.

Three export methods, picked by use case. Manual copy-paste for ad-hoc use, with a cleaning template because Notion's raw paste includes formatting artifacts and empty property tables. Notion API export for repeating workflows, with the caveat that the API returns blocks with metadata that need flattening into clean text. Notion export → local files → CLAUDE.md for Claude Code users who want persistent project context — requirements docs, ADRs, coding standards that live in Notion but need to inform every session.

The relevance hierarchy is the heart of the skill. Always include: direct context for the current task, key decisions and their rationale (Claude needs to know WHY, not just WHAT), constraints and non-negotiables. Include if space allows: related background, user preferences, historical context of what was tried before. Rarely include: full database exports (summarise instead), meeting notes older than two weeks (extract the decisions, discard the discussion), template boilerplate. Never include: empty properties, block IDs, colour codes, synced block references that should be resolved, full page histories — everything that's structural noise rather than signal.

Compression strategies handle the budget. Databases get summarised, not exported in full ("Projects DB has 12 active. The 3 relevant: [name, status, key detail]"). Meeting notes get extracted to decisions ("Meeting [date]: Decided X because Y. Actions: A (owner, date). Open: Z"). Hierarchical loading gives Claude an index of available content and lets it request specific pages on demand — instead of you guessing what to include, Claude tells you what it needs.

Two principles do the heavy lifting. Structure the context for Claude, not for humans: Notion pages are designed for human reading — toggles, columns, databases — but Claude doesn't need visual structure. It needs clear text with explicit relationships. Convert visual hierarchy to textual hierarchy. Context goes stale: project status refreshes daily, meeting notes after each meeting, reference docs weekly, strategy docs monthly. If you exported on Monday and it's now Thursday, the project context is probably out of date and Claude is now reasoning from history. The goal of the skill isn't a one-off perfect bridge — it's teaching the user to think about context preparation so they can do it themselves for any workflow.