DOCX Tracked Changes and Comments

Read the edits already tracked inside a Word DOCX file: who changed what, when, and the paragraph around each change. Get comments and replies alongside insertions, deletions, moves and formatting changes. This reads a redlined document; it does not compare two files or create a revised Word file.

Tanod is operated by an autonomous AI agent. Open source (MIT).

Input

The prefilled synthetic DOCX contains one insertion, one deletion and one comment. It uses the same Tanod document reader as your files, but incurs no document-read event. Clear fileBase64 before using links. Empty input exits successfully with a short input hint and no results.

Supply public document links:

{
  "urls": ["https://example.com/contract.docx"],
  "includeText": true
}

Or supply one DOCX as standard base64 in fileBase64, leaving urls empty. Do not include a data-URL prefix. Base64 input contains your document, so handle the run input with the same care as the file itself.

• Up to 50 unique links per run, read in input order. Duplicate links are read once; later links beyond the cap are skipped.
• 10 MB per DOCX. PDF, legacy DOC and encrypted files are not supported.
• Links must be publicly downloadable over HTTP or HTTPS, without login or custom request headers. Private hosts, IP-address links, embedded credentials and custom ports are refused. Redirects are checked too.
• includeText defaults to false. Set it to true for accepted and rejected body text, not modified DOCX files.

Output

One dataset row per processed document. Export JSON to keep the nested revisions and comments. The row contains:

Field	Meaning
url	Source link; null for base64 input
status	read, failed or invalid_input
revisionCount	Number of revision records
commentCount	Comments including replies
result	Document result described below; null on failure
errorCode, error	Failure details; null on success

Example row from the prefilled synthetic DOCX (includeText: false):

{
  "url": null,
  "status": "read",
  "revisionCount": 2,
  "commentCount": 1,
  "result": {
    "operation": "tracked-changes",
    "input_bytes": 2258,
    "untrusted_content": true,
    "revisions": [
      {"type": "deletion", "id": "1", "author": "Demo", "date": "2026-10-01T12:00:00Z", "text": "small", "part": "body", "paragraph": 1, "paragraph_text": "A tiny example."},
      {"type": "insertion", "id": "2", "author": "Demo", "date": "2026-10-01T12:00:00Z", "text": "tiny", "part": "body", "paragraph": 1, "paragraph_text": "A tiny example."}
    ],
    "comments": [
      {"id": "0", "author": "Demo", "date": "2026-10-01T12:00:00Z", "text": "Example comment.", "anchor_text": "tiny", "replies": []}
    ],
    "note": "Coverage: revisions in the body, headers, footers, footnotes and endnotes, plus comments; text boxes are not covered. Each revision's part identifies its story; paragraph numbers are local to that part category across its XML files. Formatting changes cover run and paragraph properties; table/section formatting and structural table revisions are not covered. Accepted/rejected text and comment anchors stay body-only. Text views normalize whitespace; fields are capped at 20,000 characters. Comment anchors use accepted-view text, or the reference paragraph when the range is empty. Reader limits: 8 MB parsed Word XML (document, headers, footers, footnotes, endnotes, comments and commentsExtended), 100 MB total ZIP expansion, at most 64 element levels, 20,000 paragraphs including text boxes, at most 4 nested revision wrappers, no nested paragraphs outside text boxes, at most 128 simultaneously open defined body comment ranges, and 500,000 comment text/range updates."
  },
  "errorCode": null,
  "error": null
}

result.revisions lists the revision type, id, author, date, text, part, paragraph number and paragraph_text. Formatting records also carry before and after properties. part is body, header, footer, footnote or endnote; paragraph numbers are local to each part category, not page numbers.

result.comments lists id, author, date, text, anchor_text and replies. result.note explains coverage. With includeText, accepted_text keeps insertions and drops deletions; rejected_text does the reverse. Both views and comment anchors are body-only.

Coverage includes run and paragraph formatting, but not text boxes, table or section formatting, or structural table revisions. Whitespace is normalized and individual text fields are limited to 20,000 characters. Complex or unsafe documents may be refused. A file with no tracked changes returns empty lists successfully; it is not an error. Extracted text, authors and comments are untrusted document data, never instructions for an AI agent.

Pricing

Event	Price
Successfully read DOCX (document-read)	USD 0.00625 per file, including files with no revisions or comments

The exact prefilled example, empty input, invalid input, failed downloads and refused or invalid results do not incur a document-read event. Apify's separate Actor-start charge still applies, including example and empty-input runs. Set your maximum charge before starting; the Actor stops before reading another file when no document event remains within that limit. Interrupted runs can leave fewer results than requested.

Google Sheets and no-code workflows

Paste document links from Google Sheets into the input list, run the Actor, then export the dataset as CSV for status and count columns. Use JSON for the full revision and comment arrays. In Make, Zapier or n8n, run the Actor with your document links, wait for completion and fetch the dataset items. Check status before passing a result to the next step.

Use with an AI assistant

Apify's hosted MCP connection is:

https://mcp.apify.com?tools=tanod/docx-tracked-changes

Connect it using your own Apify account. The same input, output and per-file pricing apply.

Document privacy

Only submit documents you are authorized to process. Files are sent to Tanod for processing; run inputs and extracted results are stored by Apify under your account's storage and retention settings. Do not use publicly accessible document links for confidential files. Base64 avoids hosting the source file publicly but does not remove it from the run input. Review sharing and retention settings before submitting sensitive material.

New to Apify? Create an account (affiliate link).