AI Doc Sidekick
Split large PDFs into model-sized chunks — without uploading a thing.
AI Doc Sidekick is a privacy-first PDF chunker for frontier AI models. Drop in a long document and it produces a ZIP of correctly-sized chunks plus a CLAUDE.md index file you can hand straight to Claude, Gemini, or any other long-context model.
**Your PDF never leaves the browser.** All splitting, token counting, and ZIP packaging happens client-side. The companion Worker is stateless — it serves only model metadata and never sees document bytes.
Tuned for 2-token overlap between chunks, granular by page boundary, and ready for whichever model you choose. Includes a built-in MCP (Model Context Protocol) server for agent integrations.
- 01
Client-side splitting
PDF bytes never leave the browser. The Worker is metadata-only.
- 02
Page-boundary chunks
2k-token overlap, granular by page, deterministic token counting (no AI in the splitting logic).
- 03
Drop-in for Claude
Outputs a ZIP plus a CLAUDE.md index mapping each chunk for direct use in Claude or other frontier models.
- 04
MCP server included
JSON-RPC 2.0 endpoint exposes model constraints for agentic workflows.