llamaparse-plugin vs. ocrwell

LlamaCloud document operations for Claude Code: parse, extract, classify, and split documents with LlamaIndex's hosted services, all from one plugin. Four lazy-loaded skills, each routed by name so context only enters Claude's window when needed: - llamaparse — turn PDFs, Word, PowerPoint, Excel, images, and scans into clean markdown or text via LlamaParse v2 (handles tables, multi-column layout, OCR, mixed text+images). - llamaextract — pull structured JSON out of a document against a JSON Schema you supply (invoice line items, contract terms, form fields, entities). - llamaclassify — categorize a document into one of a defined set of labels with a confidence score and a short reasoning string. - llamasplit — break a long document into typed sections with page ranges and per-section confidence (LlamaSplit v1 beta). Includes a standalone `llamaparse` CLI (`parse / extract / classify / split` subcommands plus `doctor` for preflight checks) so the same code powers scripts, CI jobs, and the Claude Code skills. R

Teaches Claude Code the OCRWell document OCR API so it generates correct integration code without having to be fed documentation. The agent learns the /v1 endpoints, the two-step upload flow (create upload → PUT to presigned URL → create document), asynchronous job polling with backoff, both structured-extraction schema formats (template and JSON Schema), the full error catalogue, rate-limit headers, and HMAC-SHA256 webhook signature verification. It also knows the quiet gotchas, like treating a 200 response carrying job.status: "failed" as a job failure rather than an HTTP success.