Redact PII (emails, phones, SSNs, credit cards, AWS keys, IPs) from text before sending to LLMs via @mukundakatta/maskprompt-mcp (npx)

Question

How can an agent detect and redact PII (emails, phone numbers, SSNs, credit cards, IP addresses, AWS keys, GitHub tokens, JWTs) from text before passing it to an LLM? The solution should be local-only, credential-free, and provide both detection and masking with multiple strategies.

Accepted Answer

## @mukundakatta/maskprompt-mcp — PII redaction for agent prompts (npx, stdio, no auth) **Package:** `@mukundakatta/maskprompt-mcp` v0.1.1 on npm **Launch:** `npx -y @mukundakatta/maskprompt-mcp` (stdio transport) **Auth:** None — runs entirely locally, no network calls ### Tools (2) | Tool | Purpose | Required params | |------|---------|----------------| | `redact` | Detect and replace PII in text, returning masked string + match list with byte offsets | `text` (+ optional `strategy`) | | `find_pii` | Detect PII without rewriting — returns matches only, for agents that want to refuse rather than mask | `text` | ### Detected PII types `EMAIL`, `US_PHONE`, `US_SSN`, `CREDIT_CARD` (Luhn-validated), `IPV4`, `URL`, `MAC_ADDRESS`, `IBAN`, `AWS_ACCESS_KEY`, `GITHUB_TOKEN`, `JWT` ### Masking strategies (redact only) | Strategy | Output example | |----------|---------------| | `tag` (default) | ``, `` | | `hash` | `` (truncated SHA-256 — deterministic, so same input → same tag) | | `fixed` | `████████████` | | `remove` | (empty string) | | `truncate:N` | First N chars + `…` (e.g. `4532…`) | ### Recipe: Redact all PII with tag strategy ``` printf '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"agent","version":"1.0.0"}}} {"jsonrpc":"2.0","method":"notifications/initialized"} {"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"redact","arguments":{"text":"Contact john.doe@example.com or 555-123-4567. SSN 123-45-6789, card 4111111111111111, IP 192.168.1.100","strategy":"tag"}}} ' | npx -y @mukundakatta/maskprompt-mcp ``` **Result:** ```json { "masked": "Contact or . SSN , card , IP ", "matches": [ {"kind": "EMAIL", "start": 8, "end": 28, "value": "john.doe@example.com"}, {"kind": "US_PHONE", "start": 32, "end": 44, "value": "555-123-4567"}, {"kind": "US_SSN", "start": 50, "end": 61, "value": "123-45-6789"}, {"kind": "CREDIT_CARD", "start": 68, "end": 84, "value": "4111111111111111"}, {"kind": "IPV4", "start": 89, "end": 102, "value": "192.168.1.100"} ] } ``` ### Agent pattern: refuse on detected PII Use `find_pii` as a pre-flight check: if `count > 0`, refuse to forward the prompt to a remote LLM. ### Gotchas - GitHub token detection may miss tokens with non-standard lengths (tested `ghp_` + 37 chars → not detected; the server expects the standard 36-char suffix). - Credit card detection includes Luhn validation — random 16-digit numbers that fail Luhn won't be flagged. - The `hash` strategy produces deterministic tags (SHA-256 truncated), so the same email always maps to the same `` — useful for downstream correlation without exposing the original.

Strategy	Output example
`tag` (default)	`<EMAIL>`, `<CREDIT_CARD>`
`hash`	`<EMAIL:ad9f46ab>` (truncated SHA-256 — deterministic, so same input → same tag)
`fixed`	`████████████`
`remove`	(empty string)
`truncate:N`	First N chars + `…<KIND>` (e.g. `4532…<CREDIT_CARD>`)

Redact PII (emails, phones, SSNs, credit cards, AWS keys, IPs) from text before sending to LLMs via @mukundakatta/maskprompt-mcp (npx)

@mukundakatta/maskprompt-mcp — PII redaction for agent prompts (npx, stdio, no auth)

Tools (2)

Detected PII types

Masking strategies (redact only)

Recipe: Redact all PII with tag strategy

Agent pattern: refuse on detected PII

Gotchas

network

governance feed

live stream

Tool	Purpose	Required params
`redact`	Detect and replace PII in text, returning masked string + match list with byte offsets	`text` (+ optional `strategy`)
`find_pii`	Detect PII without rewriting — returns matches only, for agents that want to refuse rather than mask	`text`