npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2026 – Pkg Stats / Ryan Hefner

@pii-remover/mcp-server

v0.0.5

Published

MCP (Model Context Protocol) server exposing @pii-remover/core sanitize/desanitize/analyze tools

Readme

@pii-remover/mcp-server

MCP (Model Context Protocol) server exposing @pii-remover/core as 5 tools that any MCP-compatible client can call:

| Tool | Purpose | |---|---| | sanitize | Mask PII in one text → return vault_id | | sanitize_batch | Mask many texts inside one shared vault (token dedup) | | desanitize | Restore PII tokens using vault_id | | desanitize_batch | Restore many texts from one vault | | analyze | Detect PII spans without creating a vault (no original text in response) |

Designed for Claude Desktop, Cursor, Cline, Cody, opencode, Continue, Aider, or any client that speaks MCP.

See docs/ADR/0016-mcp-server-package.md for the design rationale.

Quick start

Claude Desktop / Cursor (stdio, recommended)

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "pii-remover": {
      "command": "npx",
      "args": ["-y", "@pii-remover/mcp-server"]
    }
  }
}

Restart Claude Desktop. The 5 tools appear in the tool picker.

Streamable HTTP (remote / container scenario, opt-in)

Start the server:

pii-remover-mcp --transport http --port 8766

Client config:

{
  "mcpServers": {
    "pii-remover-remote": {
      "url": "http://localhost:8766/mcp"
    }
  }
}

Detection backend

The MCP server delegates detection to @pii-remover/core, which by default looks for a local OPF Docker backend on http://localhost:8000/redact. To run with the built-in LocalRegexBackend only (no Docker required), set backend.endpoint = "" in your config — useful for development and CI:

// .pii-remover.json
{
  "backend": {
    "type": "single",
    "endpoint": "",
    "trust_tier": "local",
    "auth": { "type": "none" }
  }
}

Pass --config /path/to/pii-remover.json to override.

CLI flags

pii-remover-mcp [--transport stdio|http] [--port N] [--host HOST]
                [--config PATH] [--max-vaults N] [--ttl-ms N]

| Flag | Default | Notes | |---|---|---| | --transport | stdio | http enables Streamable HTTP (ADR-0016 §2) | | --port | 8766 | HTTP transport only | | --host | 127.0.0.1 | HTTP transport only | | --config | env-resolved | Path to pii-remover.json | | --max-vaults | 100 | LRU pool size | | --ttl-ms | 3600000 (1h) | Vault idle TTL |

Tool surface

sanitize

Input:

{ "text": "Reach me at [email protected]",
  "vault_id": "session_..." }            // optional

Output (structuredContent):

{
  "text": "Reach me at __OPF_EMAIL_1__",
  "vault_id": "session_a1b2…",
  "token_count": 1,
  "categories": { "private_email": 1 },
  "latency_ms": 3.2,
  "backend_name": "local-regex"
}

Reuse the returned vault_id on subsequent sanitize calls to keep token dedup (same PII → same token). Pass it to desanitize to round-trip the response back to the original PII.

sanitize_batch

Same as sanitize, but takes texts: string[] and returns results: SanitizeOutput[]. All inputs share one vault.

desanitize

Input:

{ "text": "Hi __OPF_EMAIL_1__",
  "vault_id": "session_a1b2…" }          // required

Output:

{
  "text": "Hi [email protected]",
  "restored_count": 1,
  "unknown_token_count": 0,
  "partial_match_count": 0,
  "vault_id": "session_a1b2…"
}

If vault_id is missing or expired, returns isError: true with structuredContent.error_code = "vault_not_found" or "vault_expired".

desanitize_batch

Takes texts: string[] + required vault_id. Returns results: DesanitizeOutput[].

analyze

Input:

{ "text": "secret [email protected] here" }

Output:

{
  "detections": [
    { "start": 7, "end": 23, "category": "private_email", "confidence": 0.9 }
  ],
  "backend_name": "local-regex",
  "latency_ms": 0.8
}

Security note: response intentionally omits the original PII text. If you need the masked text, use sanitize instead. No vault is created.

Vault lifecycle

Per ADR-0016 §4:

  • vault_id is a server-generated opaque string, independent of MCP-Session-Id.
  • LRU eviction when pool reaches --max-vaults (default 100). Oldest entry first.
  • TTL eviction after idle period (--ttl-ms, default 1 hour). Sweeper runs every ttlMs/4.
  • All vault data is in-memory only — never persisted to disk (ADR-0003 invariant).
  • Pool is single-tenant per server process. Multi-tenant client isolation is v2.

Error semantics

Per ADR-0016 §7:

| Situation | Response | |---|---| | Schema violation (e.g., text is a number) | JSON-RPC -32602 (SDK auto) | | vault_id missing or unknown | tool isError: true, error_code: "vault_not_found" | | vault_id expired (idle past TTL) | tool isError: true, error_code: "vault_expired" | | Detection backend fail-closed | tool isError: true, error_code: "fail_closed" | | Server internal panic | JSON-RPC -32603 (SDK auto) |

Logging

logging capability declared. Use any MCP client supporting logging/setLevel to control verbosity. Logs never contain PII plaintext — only category counts, vault IDs, backend names, latencies, error classes.

In stdio mode, stdout is the JSON-RPC channel; runtime logs flow through notifications/message, never stdout. stderr is reserved for fatal boot errors only.

Build / test

bun install
bun run --filter '@pii-remover/mcp-server' typecheck
bun run --filter '@pii-remover/mcp-server' build
bun test packages/mcp-server/tests/

Single-file binary:

cd packages/mcp-server
bun run compile:linux-x64       # → dist/pii-remover-mcp-linux-x64
bun run compile:darwin-arm64    # → dist/pii-remover-mcp-darwin-arm64
bun run compile:darwin-x64      # → dist/pii-remover-mcp-darwin-x64
bun run compile:windows-x64     # → dist/pii-remover-mcp-windows-x64.exe

Programmatic use

import { createPiiRemoverMcpServer, runStdio } from "@pii-remover/mcp-server";

const server = createPiiRemoverMcpServer({
  vaultPoolOptions: {
    maxSize: 50,
    ttlMs: 30 * 60 * 1000,
  },
});

await runStdio(server);

For programmatic Streamable HTTP:

import { createPiiRemoverMcpServer, runStreamableHttp } from "@pii-remover/mcp-server";

const server = createPiiRemoverMcpServer();
const { close } = await runStreamableHttp(server, { port: 8766 });
// ... later
await close();

Limitations (v1)

  • Single-tenant per process — multiple MCP clients hitting the same server share vault state (no client identity isolation).
  • No analyze_context_risk (CloakLLM parity) — ContextAnalyzer not yet implemented in core.
  • No dispose_vault tool — vaults clean up automatically via LRU+TTL. Add an explicit tool if your usage demands it.
  • No MCP resources / prompts / sampling — those primitives are out of scope.
  • SDK v1.x stable line; v2 split-package migration planned when v2 reaches stable.

License

Apache-2.0