Unstructured Transform
ClaudecommunityTurn any document into AI-ready structured data
Unstructured
Unstructured Transform turns real-world documents into clean, structured, AI-ready data. Point it at a file — PDF, Word, PowerPoint, Excel, HTML, email, scanned images, and ~70 other formats — and it returns the content as Markdown, element-level JSON, HTML, or plain text that an agent can act on immediately. Under the hood it runs a configurable pipeline: partition documents into structured elements (titles, paragraphs, tables, lists, images) with layout and page metadata; optionally enrich them with vision-language passes (image and table descriptions, table-to-HTML, named-entity recognition, generative OCR); chunk the content for retrieval; and generate embeddings for a vector store. It can also extract structured JSON from a document against a schema you provide — or draft that schema for you. Work is submitted as an asynchronous job, and results are delivered out of band through a short-lived download link, so even large documents never overwhelm the conversation. Each request handles up to 10 files, 50 MB per file. Common uses: feed specification PDFs into a coding task, ground answers in policy or contract documents, build a searchable Q&A corpus over enterprise files, pull structured fields from forms, invoices, or contracts against a JSON schema, or prepare large document sets for a RAG pipeline. Requires a free Unstructured account. Learn more at docs.unstructured.io/transform.
Anthropic’s own published signals, snapshot of August 5, 2026; units undocumented — treat as ordinal.
Over time
Details
Tools(6)
- check_job_status
- get_job_results
- request_file_upload_url
- start_extraction_job
- start_transform_job
- suggest_extraction_schema_for_file
Data from Claude’s public directory API, snapshotted daily by the MCP App Tracker. A listing describes what the directory publishes — not an endorsement.