​

Agent-Ready Documents

Unfurl is not only for humans reading documents. It is built for a world where AI agents create, read, and act on documents alongside people.

​

The problem with traditional formats

When an AI agent needs to read a document, the format matters enormously: PDF: The agent does not read a PDF as text. It gets rasterised to images and processed via vision — expensive, slow, and lossy. Tables get misinterpreted. Headers get confused with body text. Footnotes get lost. DOCX: Better — the XML can be parsed to extract text. But the structure is verbose, formatting metadata creates noise, and round-tripping (agent reads → agent writes → human reads) is fragile. HTML: Native to LLMs. An agent reads HTML the same way it thinks — as structured text. No conversion step. No fidelity loss. No inflated token cost for vision.

​

Why this matters

An agent working with a Unfurl document:

• Reads with full fidelity — tables, headings, lists, preserved
• Costs less per read — no vision tokens, no binary parsing overhead
• Produces better output — decisions based on accurate data, not hallucinated table structures
• Gets live data — a live link returns the current version, not a stale snapshot

​

Live context, not snapshots

When you attach a PDF to an agent’s context, it is frozen in time. The moment the source document changes, the agent works with outdated information. A Unfurl link is different. It is a live endpoint. When the document updates, any agent that reads via get_document (or your integration) sees the current version. No re-uploading. No re-attaching.

​

The MCP loop

This loop runs continuously. The document is never “done” — it is a living artifact in an ongoing workflow.

​

Built for the agent era

Agents are becoming part of your environment — assigned tasks, tagged in comments, reading and writing next to humans. Unfurl is the document layer built for that world.