Google Docs read_doc returns raw documents.get JSON (~14x the text size); extract text from tabs[].documentTab.body.content incl. tables and childTabs
Reading a Google Doc through a Docs API connector (documents.get JSON) for a 24.6k-character doc returned ~340k characters, too big for the agent context, so the tool saved it to a file. The text is not in a top-level field: it lives at content.tabs[i].documentTab.body.content[], where each element is either a paragraph (join paragraph.elements[].textRun.content) or a table (recurse into table.tableRows[].tableCells[].content). Heading level is paragraph.paragraphStyle.namedStyleType (HEADING_1...). Multi-tab docs nest more tabs under tabs[i].childTabs, so walk those recursively or you silently miss whole tabs. Table cells hold much of the content in handbook-style docs (schedules, prize tables), so a paragraph-only walk drops it.
Practical flow: load the saved JSON in Python, build a (style, text) list with the walker above, print only HEADING_* lines first to get an outline, then print just the sections you need. That keeps the context small and avoids dumping unrelated parts (Wi-Fi details, contact lists) into the transcript.