PDF metadata, in your AI tools.

When the user asks to inspect a PDF, use PDFCheck’s analyze_pdf tool with file_base64 and filename. Read and base64-encode the file using your client’s file tools. Do not pass a local file path to the remote server.

Read the JSON result from content[0].text. Report the filename, page count, PDF version, author, creator, producer, and dates when present. Explain missing values. Distinguish returned facts from your interpretation.

For analysis created with your account token, use get_analysis with the returned data.id to revisit results. Use list_analyses for account history and check_usage for the current allowance. Retain the initial response from anonymous analysis; it cannot be retrieved later through MCP.

Do not infer AI authorship, document authenticity, signature trust, standards compliance, or accessibility from metadata. Other PDF operations need separately available tools.

When the user asks to inspect PDF metadata, POST the file to /api/v1/pdf/analyze as multipart/form-data using the field file. Use the configured bearer token for account access. Never include the token in your response.

Read data.metadata.basic for page count and PDF version, data.metadata.document for document fields, and data.metadata.dates for dates. Use data.id to retrieve the saved analysis from GET /api/v1/pdf/{token}.

For text extraction, use POST /api/v1/pdf/extract-text. For Markdown, use POST /api/v1/pdf/to-markdown. Both use the multipart pdf field; consult their documentation for limits and response formats.

Handle unsuccessful HTTP responses and quota errors before reporting results. Metadata is editable and cannot prove AI authorship or authenticity.

# PDF analysis

Requires: PDFCheck MCP tools or REST API access, plus a file-reading tool.

1. Read the PDF and submit it to analyze_pdf (base64 and filename) or POST /api/v1/pdf/analyze (multipart file).
2. Report filename, page count, PDF version, document fields, and dates returned by the service.
3. Mark missing fields as unavailable. Metadata values are editable and are not authenticity evidence.
4. Include the returned analysis ID. Account-created analyses can be retrieved later; for anonymous MCP analysis, retain the initial response.
5. If a request fails, report the error and a concrete next step without inventing results.

# PDF generation and inspection

Requires: a local PDF-generation library and PDFCheck analysis access. PDFCheck does not expose a PDF-generation API.

1. Generate the document with the tools available in your environment.
2. Open the rendered PDF and review page breaks, text, and images.
3. Submit the PDF to PDFCheck for metadata analysis.
4. Check the returned page count, document fields, and PDF version against the intended output.
5. Use separate specialist validation when accessibility or standards compliance is required.

# PDF review

Requires: PDFCheck for metadata; separate tools for structural or formal compliance checks.

1. Inspect metadata with PDFCheck.
2. Use the browser PDF structure checker when appropriate. It is not a public MCP tool.
3. For PDF/A, PDF/UA, accessibility, or signature trust, use suitable specialist validators and manual review.
4. Report each check, its result, and its scope. Never turn metadata indicators into a compliance certificate.

# Document workflow

Requires: PDFCheck REST API plus your own storage, routing, and review tools.

1. Analyze metadata using POST /api/v1/pdf/analyze.
2. Extract text or Markdown using the corresponding REST endpoint if needed.
3. Handle empty output, failed requests, and daily limits.
4. Pass the returned data to your routing rules and flag uncertain cases for review.
5. Use separate tools for compression, merging, or signing. Do not call browser upload endpoints as if they were public API methods.