Harness Score L0 (Unharnessed): measures AI-assisted development harness maturity with harness-score

DevUtils MCP Server

36 everyday developer tools for any MCP-compatible AI assistant. Hashing, encoding, UUID generation, JWT decoding, JSON formatting, network tools, text utilities, and more — all local, no external APIs.

Think of it as busybox for developer tools — small, essential, and always useful.

Also available as a plugin: devutils-cursor-plugin — one-click install for Cursor and Claude Code.

Why?

Every developer needs to hash strings, encode/decode data, generate UUIDs, decode JWTs, format JSON, calculate CIDR ranges, and convert timestamps every day. DevUtils MCP Server brings all of these tools directly into your AI assistant — works with Claude, Cursor, VS Code, Windsurf, and any other MCP-compatible client.

Installation

Prerequisite: Node.js 18+ (Node 22 recommended). Verify with node -v.

One-click / plugin

Use the Add to Cursor / VS Code badges above, or install the plugin:

/plugin marketplace add paladini/devutils-cursor-plugin
/plugin install devutils-mcp@devutils-cursor-plugin

Cursor: Settings → Customize, or add from GitHub paladini/devutils-cursor-plugin.

npx (no install)

npx -y devutils-mcp-server

npm

npm install -g devutils-mcp-server
devutils-mcp-server

Docker

# Published image (when available)
docker run -i --rm ghcr.io/paladini/devutils-mcp-server

# Or build locally
docker build -t devutils-mcp-server .
docker run -i --rm devutils-mcp-server

# Or with Compose
docker compose build
docker compose run --rm -i devutils-mcp

Official MCP Registry

Listed as io.github.paladini/devutils-mcp-server. Search for io.github.paladini/devutils.

GitHub Packages

Releases are dual-published to npm and GitHub Packages as @paladini/devutils-mcp-server (see .github/workflows/release.yml). Prefer the public npm package for most installs; use GitHub Packages when you already authenticate against npm.pkg.github.com.

# After authenticating to npm.pkg.github.com for the @paladini scope:
npx -y @paladini/devutils-mcp-server

Smithery

Also available on Smithery.

Client setup

Cursor

One-click: use the Add to Cursor badge at the top of this README.

Plugin (recommended): Install DevUtils MCP from Cursor Settings → Customize.

Manual: Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "devutils": {
      "command": "npx",
      "args": ["-y", "devutils-mcp-server"]
    }
  }
}

Claude Desktop

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

{
  "mcpServers": {
    "devutils": {
      "command": "npx",
      "args": ["-y", "devutils-mcp-server"]
    }
  }
}

Or with Docker:

{
  "mcpServers": {
    "devutils": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "ghcr.io/paladini/devutils-mcp-server"]
    }
  }
}

Claude Code

/plugin marketplace add paladini/devutils-cursor-plugin
/plugin install devutils-mcp@devutils-cursor-plugin

VS Code (GitHub Copilot)

One-click: use the VS Code badge at the top, or add to .vscode/mcp.json / user settings:

{
  "servers": {
    "devutils": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "devutils-mcp-server"]
    }
  }
}

Windsurf

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

{
  "mcpServers": {
    "devutils": {
      "command": "npx",
      "args": ["-y", "devutils-mcp-server"]
    }
  }
}

Docker MCP Toolkit (Docker Desktop)

If this server is available in the Docker MCP Catalog:

1. Open Docker Desktop → MCP Toolkit
2. Search for DevUtils
3. Click Enable

Local development

npm install
npm run dev

Available tools (36)

Hash (6)

Encoding (8)

Generators (4)

JWT (2)

Formatters (3)

Converters (5)

Network (2)

Text (6)

Architecture

src/
├── index.ts          # MCP server entry point (stdio transport)
└── tools/
    ├── hash.ts       # Cryptographic hash functions
    ├── encoding.ts   # Encode/decode utilities
    ├── generators.ts # ID and password generators
    ├── jwt.ts        # JWT decode and validation
    ├── formatters.ts # JSON formatting and querying
    ├── converters.ts # Data type and unit converters
    ├── network.ts    # Network calculation utilities
    └── text.ts       # Text analysis and manipulation

Tech stack: TypeScript + Node.js 22 · @modelcontextprotocol/sdk · bcryptjs · nanoid · zod

Zero external API dependencies. All tools run locally with no network calls.

Docker

The image uses a multi-stage build for minimal size:

1. Build stage: Compiles TypeScript on Node 22 Alpine
2. Runtime stage: Runs compiled JS on Node 22 Alpine as a non-root user

docker build -t devutils-mcp-server .

# Smoke-test with an MCP initialize request
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0.0"}}}' | docker run -i --rm devutils-mcp-server

FAQ & design philosophy

Why MCP, and not just a library?

Valid criticism: If you're writing Python scripts and need to hash something, hashlib is 2 lines of code. Why run MCP overhead?

Answer: This server is optimized for AI agents in multi-step workflows, not programmers writing code:

1. AI hallucination cost >> MCP overhead — An AI model spending 50ms calling an MCP tool (vs. 1ms library call) is negligible when the alternative is the model making up a hash or using the wrong encoding.
2. Reliable tool semantics — MCP enforces strict tool contracts. For example, jwt_decode always returns human-readable dates with timezone support.
3. Universally accessible — Any MCP-compatible client can use these tools. A Python library only works if your agent is Python-based.
4. Multi-tenant safety — MCP provides explicit tool whitelisting with input validation.

When to use DevUtils versus alternatives

Use DevUtils if:

• You're using Claude, Cursor, VS Code Copilot, Windsurf, or any MCP-compatible AI assistant
• You want reliable, validated utility operations in AI workflows
• You need 36 tools in one package

Don't use DevUtils if:

• You're writing regular application code (use native libraries)
• You need extreme performance (direct library calls are faster)
• Your AI client does not support MCP

Design philosophy

• Small & focused: 36 utilities, zero external APIs, ~50MB container
• Security-first: Non-root user, Alpine Linux, minimal attack surface
• AI-friendly: Consistent naming (<domain>_<operation>), strict schemas, human-readable outputs
• Client-agnostic: Works with any MCP-compatible client via stdio transport

Available on

Contributing

Questions and ideas: GitHub Discussions

Security reports: see SECURITY.md.

1. Fork the repository
2. Create your feature branch (git checkout -b feat/amazing-tool)
3. Commit your changes (git commit -m 'feat: add amazing tool')
4. Push to the branch (git push origin feat/amazing-tool)
5. Open a Pull Request

See CHANGELOG.md.

License

MIT © Fernando Paladini