See what a document may show the machine.
Inspect a file before copying or extracting its contents into an agent. The file stays in your browser; this scanner is deliberately not a remote API or MCP tool.
Scan before ingestion
Open the Document Scanner and select a PDF, DOCX/DOCM, HTML, TXT, Markdown, CSV, JSON, or XML file up to 25 MB. File contents are processed locally in the browser and are not sent to the ScrubMyText API.
Look beyond visible paragraphs
- Hidden or invisible text and suspicious instruction patterns
- Likely secrets and encoded content
- Document comments, metadata, macros, scripts, and annotations
- PDF text geometry, HTML hidden content, and mixed-script signals
Quarantine or extract deliberately
If a finding is unexpected, stop and review the source. If the document is expected, extract only the content your agent actually needs, then run that text through the Text Safety workflow before model ingestion.
Do not auto-ingest an unexpected finding
Require review whenever the scanner reports hidden content, suspicious instructions, secrets, active content, or unexplained metadata. A clean result reduces uncertainty but is not proof that the document is safe.
Why this workflow is browser-only
The scanner's privacy advantage comes from keeping the full file on the user's device. There is currently no document-upload REST or MCP endpoint. Agents that need automation should extract permitted text in their own trusted environment and send only that text to prepare_for_agent.
Known limits
- No OCR: text contained only in images can be missed.
- Prompt-injection findings are heuristic review signals, not a safety verdict.
- Encrypted, malformed, or unusual files may not expose all content.
- The tool does not authorize a document or verify its sender.