Skip to content
MCP server

Deliverd Screen

Local MCP server that redacts personal data and secrets in text before it reaches a model

Runs locallyStatus and history
Install
Pricing
Open source
Platforms
CLI, API
Runs
Locally
Version
0.3.0
Listing updated
2 days ago

About Deliverd Screen

Deliverd Screen is an open-source MCP server for security. It runs locally on your machine and is started by your AI client.

The MCP server exposes one tool, screen_text, from the open-source Deliverd Screen library. A coding agent can call it to check text before sending it on, and the tool never returns the values it replaced. The library behind it gives one of four verdicts: allow, redact, ask or deny. By default a model may receive internal data but no personal data, secrets are refused outright, and special-category terms such as health or religion are flagged for a person to decide.

Detectors cover email addresses, phone numbers, IBANs, Luhn-checked card numbers, UK National Insurance, Irish PPS and US Social Security numbers, IP addresses, dates of birth, API keys and tokens, private keys, passwords in connection strings and .env files, plus your own confidential terms. Redacted values become placeholders such as [EMAIL_1], and restore() puts the originals back in the model's answer. Names need an extra detector, and adapters read a Hugging Face token classifier or a Presidio analyzer.

It runs locally with no account, key or network call. The same checks are available as TypeScript, Python and Rust packages, a Vercel AI SDK middleware and a CLI that can fail CI when a fixture contains customer details. The MCP package is published as @deliverd/screen under the MIT licence.

Install

Run in a terminal

claude mcp add deliverd-screen -- npx -y @deliverd/screen

Add to claude_desktop_config.json

{
  "mcpServers": {
    "deliverd-screen": {
      "command": "npx",
      "args": [
        "-y",
        "@deliverd/screen"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "deliverd-screen": {
      "command": "npx",
      "args": [
        "-y",
        "@deliverd/screen"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "deliverd-screen": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@deliverd/screen"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "deliverd-screen": {
      "command": "npx",
      "args": [
        "-y",
        "@deliverd/screen"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "deliverd-screen": {
      "command": "npx",
      "args": [
        "-y",
        "@deliverd/screen"
      ]
    }
  }
}

Tools this server provides (1)

Taken from the project documentation; the running server may differ.

screen_text
Checks text for personal data and secrets and returns a verdict without the values it replaced.

What it does

screen_text tool
One MCP tool for Claude Code, Cursor or Codex that screens text and never returns replaced values.
Reversible redaction
Replaces values with stable placeholders and restores them in the model's answer.
Four verdicts
Returns allow, redact, ask or deny depending on what the model is approved to receive.
Secret detection
Spots API keys for several providers, JWTs, private keys and passwords in connection strings.
Custom confidential terms
Lets you add your own words to the list of things to catch.
Pluggable name detection
Adapters accept a Hugging Face token classifier or a Presidio analyzer for people's names.
CI and pre-commit check
A command exits with failure when files such as eval sets contain personal data.

Good for

  • Strip emails and IBANs from a support prompt before sending it to an LLM
  • Stop an API key pasted into a chat from reaching a model
  • Fail a build when an eval fixture contains a customer's details
  • Let a coding agent check a draft before it is shared externally
  • Wrap a Vercel AI SDK model so every conversation is screened

Pricing

Open source

Library is MIT-licensed; Data Gate is part of paid Deliverd plans

Prices change. Check deliverd.dev before you buy.

Deliverd Screen FAQ

Does my text leave my machine?

Not with the library, which makes no network call and needs no account or key. The optional Microsoft Purview check sends text only to your own Microsoft tenant.

Which languages are supported?

TypeScript, Python and Rust packages share the same cases, so they find the same things.

Can it find people's names?

Not by pattern. You plug in a model-based detector such as a Hugging Face token classifier or a Presidio analyzer.

What licence is it under?

MIT, with no dependencies.

How is it different from Data Gate?

Data Gate is a paid Deliverd feature that uses the same detectors, sends only counts, and routes ask decisions to a person. The library alone cannot ask anyone.

Is Deliverd Screen free?

Yes. Deliverd Screen is open source, so you can use and self-host it for free.

How do I add Deliverd Screen to Claude Code?

Run this command in your terminal: claude mcp add deliverd-screen -- npx -y @deliverd/screen

What are the best alternatives to Deliverd Screen?

The closest alternatives to Deliverd Screen are REA, Claude Code Ultimate Guide, Agent Roast, Midplane and Leash.

Alternatives to Deliverd Screen

  1. REA

    MCP server

    CLI and MCP server that lets coding agents reverse engineer apps and binaries

    Runs locally
    • Open source
    • 464K GitHub stars
    • Developer tools
  2. MCP server for searching the Claude Code Ultimate Guide from any MCP client

    Runs locally
    • Open source
    • 6.5K GitHub stars
    • AI models and LLM tooling
  3. Agent Roast

    MCP server

    Offline review that flags agent changes which weaken tests, assertions or CI

    Runs locally
    • Open source
    • 4 Hacker News points
    • Developer tools
  4. Midplane

    MCP server

    Postgres gateway for AI agents with table grants, column masking and approval for writes

    Runs locally
    • Open source
    • 5 GitHub stars
    • Databases
  5. Leash

    MCP server

    Review what your AI touched, approve risky actions and undo supported changes

    • 1 tool
    • Remote
    • Free
    • 1 GitHub star
    • Security