mcpField
R
redactpdf

redactpdf

Local
github.com/CloudCorpRecordsSecurity0Updated 1w ago

About redactpdf

redactpdf is a CLI and MCP server for removing sensitive text from local PDFs. It detects common identifiers and custom terms, writes a separate redacted copy, and checks that matched text is no longer extractable.

Key features

  1. 01Detects SSNs, cards, emails, and phones
  2. 02Matches custom terms case-insensitively
  3. 03Previews matches without writing
  4. 04Creates a separate redacted PDF
  5. 05Verifies removed text is unextractable

Use cases

  1. 01Checking a PDF for sensitive text before sharing
  2. 02Redacting identifiers from a local report
  3. 03Removing specified terms from a PDF copy

MCP support

The MCP server exposes `check_pdf` to report detected matches and `redact_pdf` to write and verify a redacted copy of a local PDF.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"cloudcorprecords-redactpdf": {
"command": "npx",
"args": [
"-y",
"@reneturcios/redactpdf"
]
}
}
}
Run commandnpx -y @reneturcios/redactpdf

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npm@reneturcios/redactpdf0.1.0
ocighcr.io/cloudcorprecords/redactpdf:0.1.0

Tools

2

Saved tool list · updated 1w ago

ParameterTypeRequiredDescription
pathstringYesAbsolute path to a PDF on this machine
termsarrayNoLiteral strings to match, case-insensitive
detectarrayNoAny of ssn, card, email, phone. Omit for all

Details

From the project README

redactpdf

Sharing a PDF usually means trusting that whatever you blacked out on screen is actually gone. Often it is not: the text is still sitting in the file, ready to be copied out. redactpdf takes the opposite approach. It looks for sensitive text in a local PDF's text layer, writes a copy in which every matched character has been physically replaced, and then re-reads its own output to prove that nothing it was asked to remove can still be extracted. If the proof fails, you get an error instead of a file.

Four kinds of sensitive text are detected. Social Security numbers must satisfy the SSA's own validity rules (area 001-899 except 666, group 01-99, serial 0001-9999), so random digit groups in that shape are not flagged. Payment-card numbers are digit runs of 13 to 19 digits, allowing single spaces or dashes between digits, and must pass the Luhn check. Email addresses and North-American phone numbers — (ddd) ddd-dddd, ddd-ddd-dddd, ddd.ddd.dddd, or ddd ddd dddd, with an optional +1 prefix — round out the built-in categories. On top of those, terms lets you name literal strings of your own, matched case-insensitively and counted under their own key. Omit detect to run all four categories.

Detection runs per extracted string — the string operands of Tj and TJ show operators, in plain or FlateDecode content streams — so a match that is split across two string operands is not seen. That is a stated limitation of the text-layer model this tool uses, not a silent one.

check_pdf reports what would be removed without writing anything. redact_pdf writes the redacted copy — the input file is never modified — replacing each matched character with # while preserving string lengths, then verifies the result. Documents run to 10 pages free; past that, both tools ask for a licence. The licence check in this build is a format check — a token of 16 to 512 characters from [A-Za-z0-9_-] plus . separators counts as well-formed, supplied via the license argument or the BLACKOUT_LICENSE environment variable — and the over-limit error distinguishes a missing licence (LICENSE_REQUIRED) from a malformed one (LICENSE_INVALID). Encrypted PDFs, PDFs built on xref or object streams, and malformed files are reported as unsupported rather than passed over in silence.

Both tools answer with a one-line summary, a blank line, and a fenced JSON block carrying path, outputPath (null for check_pdf), pages, counts, extractableChars — for check_pdf, the characters that fall inside a detected match in the input; for a successful redact_pdf, 0 by construction — and licenseState.

CLI

npx @reneturcios/redactpdf check_pdf '{"path":"/abs/path/report.pdf"}'
npx @reneturcios/redactpdf redact_pdf '{"path":"/abs/path/report.pdf","terms":["falcon"]}'
npx @reneturcios/redactpdf --version

redact_pdf writes <input>-redacted.pdf (the input path with its .pdf suffix replaced) unless output_path says otherwise; the output may not be the input itself, and an existing output is only replaced when overwrite is true.

Exit codes: 0 on success, 1 on a runtime error (the tool's error text is printed), 2 on a usage error.

MCP server

npx -y @reneturcios/redactpdf mcp

MCP client config:

{
  "mcpServers": {
    "redactpdf": {
      "command": "npx",
      "args": ["-y", "@reneturcios/redactpdf", "mcp"]
    }
  }
}
Tools
  • check_pdf — Report what would be redacted from a local PDF — per-category match counts — without writing anything.
    • path (string, required) — absolute path to a PDF on this machine
    • detect (string array, optional) — any of ssn, card, email, phone; omit for all
    • terms (string array, optional) — literal strings to match, case-insensitive
  • redact_pdf — Write a redacted copy of a local PDF with detected sensitive text permanently removed and verified gone. Never modifies the input.
    • path (string, required) — absolute path to a PDF on this machine
    • output_path (string, optional) — defaults to <input>-redacted.pdf; must not be the input
    • detect (string array, optional) — any of ssn, card, email, phone; omit for all
    • terms (string array, optional) — literal strings to match, case-insensitive
    • overwrite (boolean, optional, default false) — allow replacing an existing output
    • license (string, optional) — Pro licence token; falls back to BLACKOUT_LICENSE

Docker

docker run -i ghcr.io/cloudcorprecords/redactpdf
docker run --rm -v "$PWD:/w" ghcr.io/cloudcorprecords/redactpdf check_pdf '{"path":"/w/report.pdf"}'

Development

npm install
npm run build   # tsc -> dist/
npm test        # tsx --test test/*.test.ts (pure local, no network)
npm run dev     # run the CLI from source

License

MIT — Copyright (c) 2026 Rene Turcios

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.