Metadata Cleaner — inspect and remove metadata from PDF and Word

Inspect a PDF or Word DOCX, review the findings, then create a separately verified cleaned copy. Inspection is the default and does not change your file. This cloud Actor uses a Python port of Metadata Cleaner's services; it is a separate delivery channel from the ChatGPT, Codex, Kimi and MiniMax editions. Their purchases and balances do not automatically apply here.

Quick start

Run with one file source:

{"operation":"inspect","documentUrl":"https://your-file-host.example/document.docx"}

Supply a direct public HTTPS download URL without redirects, or use documentBase64 instead. Input values are marked secret in Apify's input schema. URLs requiring cookies, custom headers, private network access, redirects or ports other than 443 are rejected. Files are limited to 25 MiB; expanded DOCX packages are limited to 100 MiB and 4,096 ZIP entries.

Inspection returns finding titles and risk levels, a document_sha256, and requires_content_confirmation. It does not echo author names or other extracted private values.

For metadata-only cleaning:

{"operation":"clean","documentUrl":"https://your-file-host.example/document.pdf"}

When inspection reports tracked changes, hidden text or custom XML, review the effects below and resubmit the same document with explicit approval and its inspection hash:

{"operation":"clean","documentUrl":"https://your-file-host.example/document.docx","confirmContentChanges":true,"inspectedSha256":"COPY_THE_64_CHARACTER_HASH_FROM_INSPECTION"}

Without matching approval, the result is confirmation_required and no file is released. Changed file bytes invalidate the approval.

Cleaning scope

DOCX: supported author/editor and document properties, comments, tracked changes, explicitly hidden runs, custom XML and revision identifiers. Accepted insertions remain; deletions and hidden runs are removed. Comments/custom XML parts and their relationships/content-type references are repaired. Images remain unchanged, including any metadata inside those images.

PDF: supported document information, catalog XMP, markup annotations, catalog embedded-file/JavaScript entries and document/page actions. Links and form widgets are preserved; forms are not flattened. Hidden layers and form content may remain. This is not redaction or a guarantee of anonymization. Visible content, links, images and form answers can still identify people. Review the returned document before sharing it.

Unsupported, encrypted or incomplete results are refused. A re-scan must complete all supported categories with no high-risk finding remaining before a cleaned artifact is released. Notes identify retained features. Already-clean files are returned byte-for-byte, with the report explicitly marking that no cleaned-copy re-scan was necessary.

Output and download

A report is saved to the default dataset and the key-value store's OUTPUT record. Statuses: inspected, confirmation_required, cleaned, cleaned_with_notes, or could_not_clean. Only the two cleaned statuses have a file. Ordinary document refusals produce a successful Actor run containing could_not_clean; API callers must inspect the report status.

cleaned_file identifies the store and record (cleaned.pdf, cleaned.docx, or document.* when unchanged). Cloud reports contain the standard record endpoint. Fetch it with your Apify API token in the Authorization: Bearer ... header. Tokens are never embedded in result URLs. No public sharing URL is created and no expiring-link guarantee is made. In local runs, the file is in local key-value storage.

Files are uploaded and processed on Apify infrastructure. Input and output records remain subject to your Apify storage retention and access settings; this Actor does not automatically delete them. Delete the run's input, dataset and output storage when no longer needed, and do not share storage containing private documents. Inspection hashes and category reports are also stored. No document is sent to an LLM or the separate Metadata Cleaner service.

Pricing and MCP

Per successfully cleaned document: $4 on Apify Free; $0.02 on paid Apify plans. Inspection, confirmation requests, refused files, and already-clean files have no cleaning-event charge. Platform costs are included in the displayed event price; your Apify subscription is separate. One input file per run; use repeated runs for multiple files. Set your maximum event budget to at least your plan’s per-document price. Your Apify subscription determines the tier; this Actor does not sell a separate subscription. The Actor honors the run's event spending limit and charges the document-cleaned event only after a verified file is stored. Existing ChatGPT/Codex/Kimi/MiniMax balances are separate.

Works with MCP through Apify's MCP server. Connect your compatible client to https://mcp.apify.com, authorize your Apify account, and select myworkingmemory/document-metadata-cleaner. Ask it to inspect your PDF or DOCX, explain any proposed content changes, and clean only after your approval. Supply a downloadable public HTTPS URL or base64; local disk paths are not cloud-upload instructions. Retrieve the result using your Apify storage access. This Actor uses limited permissions; it does not request access to unrelated Actors or account storage.

Documentation: https://docs.apify.com/mcp

Connect through MCP

Add this server URL to a compatible MCP client such as Claude or Cursor, then authorize your Apify account. It preloads this Actor as a tool:

Apify MCP — PDF & Word Metadata Cleaner

https://mcp.apify.com/?tools=myworkingmemory/document-metadata-cleaner

Connecting does not start a run. Ask your agent to review the input and pricing before running; the same prices, input requirements and limits apply. See the Apify MCP setup guide.

More tools from Myworkingmemory

• Email Verifier — Verify email lists or Apify datasets; limited keyless pilot and BYOK options.
• Shopify Product Scraper — Extract Shopify product prices, inventory and variants.
• Curated Finds — Discover curated GitHub tools and Hugging Face models.

Each Actor has its own inputs and pricing; follow its README before running. Browse all our Actors.