PDF Compare: Text Changes + Annotated Diff PDF (MCP server)

A deterministic PDF comparison tool for AI agents, served as an MCP server. There is no LLM inside, so the same pair of files gives the same output every time. It runs on CPU and does not call any paid API.

Use it for: contract redlines when the other side did not track changes, policy and terms updates, new versions of forms and standards, price list changes, regulatory documents, and checking that a "final" PDF matches the approved draft.

Why agents pick it: asking an LLM "what changed?" across two long PDFs misses edits and invents others. This tool diffs the actual words, so every reported change is really there and nothing is skipped. It returns a structured list your agent can reason over, plus an annotated PDF a human can check in a minute.

Tool

What your agent gets back

compare_pdfs

changes: each with type (added / removed / changed / moved), old_text, new_text, old_pages, new_pages and context_before. A summary with counts and word totals, and identical_text when nothing differs. diff_pdf: a summary page listing every change, followed by the new file with green boxes (added), amber boxes (changed, with the old wording in a note) and red markers where text was removed (removed wording in a note).

How it stays accurate: two-column pages are read column by column; a line-level diff finds changed regions, then a word-level diff inside each region pins the exact words, so text that only re-wrapped does not show as an edit. Passages that were cut from one place and pasted in another are reported once as moved. Scanned pages are OCR'd.

Connect

MCP endpoint (streamable HTTP):

https://doubleaigents--pdf-compare-mcp.apify.actor/mcp
Authorization: Bearer <APIFY_TOKEN>

Claude Desktop / Cursor style config:

{ "mcpServers": { "pdf-compare": {
    "url": "https://doubleaigents--pdf-compare-mcp.apify.actor/mcp",
    "headers": { "Authorization": "Bearer <APIFY_TOKEN>" } } } }

Tool inputs

compare_pdfs(old_file_url | old_file_base64, new_file_url | new_file_base64, ignore_case=false, lang="eng", include_inline_pdf=true)

• Both files must be PDFs (public http/https URL or base64), up to 50 MB and 500 pages each.
• The diff PDF and the full change list come back as key-value store URLs. A diff PDF under 8 MB also comes back inline as base64 unless include_inline_pdf is false.

Pricing (pay per event)

Event	USD
page-compared: each page compared, counted as the page count of the longer file	0.01

You are charged only after a call completes. If your run's maximum charge cannot cover the page count, the call is refused before any work is done.

Limits worth knowing

• It compares text, not images, colours or fonts. A changed logo or a changed figure in a chart is not reported.
• Heavily re-laid-out documents (a table rebuilt as paragraphs, content moved between pages) produce more, smaller changes. Every one is real, but some may be layout rather than wording.
• Boxes are drawn on unrotated pages. On rotated pages the change notes are still added.
• Scans are OCR'd with Tesseract (English by default); OCR noise on poor scans shows up as small changes.

Batch mode

You can also run the Actor normally with oldFileUrl and newFileUrl. It writes one dataset item with the counts, the first 1,000 changes, and links to the diff PDF and the full changes JSON.

Who runs it

Double AIgents Ltd, UK (company no. 17513072). Your input files are processed for the call. Outputs sit in the run's Apify storage, which Apify removes under its data-retention policy. Need volume pricing, a private deployment or a custom document pipeline? https://www.doubleaigents.com

More tools from Double AIgents

• PDF Tables to Excel + True PII Redaction — tables from PDFs and scans to XLSX with every figure double-read, and true redaction of personal data with a verification log.
• Invoice & Receipt to JSON + Excel — supplier, dates, VAT number, line items and net/VAT/gross, with the totals reconciled.
• Bank Statement PDF to CSV/Excel — every transaction, with running balances and opening/closing totals verified.