# Connect a Local Developer Toolbox to Any MCP Assistant > Source: > Published: 2026-08-28 12:35:52+00:00 If an AI assistant can write code but cannot reliably hash a value, inspect a JWT, validate JSON, or calculate a CIDR range, you have a small but recurring reliability problem. Asking the model to do those jobs from memory adds an unnecessary interpretation step. [DevUtils MCP Server](https://github.com/paladini/devutils-mcp-server) packages 36 everyday developer utilities behind the [Model Context Protocol](https://modelcontextprotocol.io/). The server runs locally over standard input and output, so an MCP-compatible client can call explicit tools instead of guessing an operation. This tutorial connects the released `1.1.0` package, verifies the protocol handshake, and shows how to choose a useful tool without treating the server as a replacement for application libraries. Install Node.js 18 or newer, add the server command to your MCP client's configuration, restart the client, and ask it to use a tool such as `json_validate` , `jwt_validate` , or `cidr_calculate` . The smallest configuration is a command plus the package name: ``` { "mcpServers": { "devutils": { "command": "npx", "args": ["devutils-mcp-server"] } } } ``` The released package declares Node.js `>=18` . The repository's current default branch has moved ahead to `1.1.1` , so the commands and behavior in this article target the immutable `v1.1.0` release and the npm `latest` package that was verified during research. You need: `npx` and download the public npm package on first use.No API key, account, database, or external service is needed for the local server. The MIT-licensed repository lists Claude Desktop, Cursor, VS Code, Windsurf, Docker, and other MCP-compatible clients as possible consumers. Their configuration file locations differ, but the server entry is the same. The release README documents an `npx` path that does not require a global installation: ``` npx devutils-mcp-server ``` For an automated setup where accepting the package prompt must be explicit, use npm's yes flag while keeping the package name unchanged: ``` npx -y devutils-mcp-server ``` The public npm registry reported `1.1.0` as the latest version when this tutorial was checked. If reproducibility matters more than following the moving `latest` tag, pin the version: ``` npx -y devutils-mcp-server@1.1.0 ``` That distinction matters here because the GitHub default branch already contains unreleased `1.1.1` metadata. A tutorial should not silently mix the two. For Claude Desktop on Windows, the release README points to `%APPDATA%\Claude\claude_desktop_config.json` . Add the `devutils` entry inside the existing `mcpServers` object. Do not replace other servers that are already configured. ``` { "mcpServers": { "devutils": { "command": "npx", "args": ["-y", "devutils-mcp-server@1.1.0"] } } } ``` The same command shape works in the corresponding Cursor and Windsurf configuration files. VS Code uses a `servers` object instead: ``` { "servers": { "devutils": { "type": "stdio", "command": "npx", "args": ["-y", "devutils-mcp-server@1.1.0"] } } } ``` Restart the client after saving the file. The client starts the process and speaks MCP over stdio. You should not send ordinary log messages to stdout when building a similar server because stdout carries the protocol stream. DevUtils writes its startup message to stderr and returns MCP responses on stdout. You can test the process independently of an AI client by sending an MCP `initialize` request. This uses only JSON-RPC and does not expose a secret or call a remote API: ``` printf '%s\n' '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"smoke-test","version":"1.0.0"}}}' | npx -y devutils-mcp-server@1.1.0 ``` The response should be JSON-RPC with a `serverInfo` object whose name is `devutils-mcp-server` . The released server reports version `1.1.0` and advertises tool support. In a PowerShell environment, the same test can use a here-string piped to the command: ``` php $request = '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"smoke-test","version":"1.0.0"}}}' $request | npx -y devutils-mcp-server@1.1.0 ``` If the command appears to hang, check that the client or shell is closing stdin after the request. A long-running stdio server normally keeps waiting for more messages. Once the client discovers the server's tools, start with an operation whose expected shape is easy to inspect. For example, ask the assistant to validate this payload with `json_validate` : ``` { "name": "Ada", "roles": ["reviewer", "maintainer"] } ``` The server's formatter tools include JSON validation, formatting, minification, and dot-notation queries. Other useful first calls include: `hash_sha256` for a deterministic digest of a supplied string.`cidr_calculate` for network, broadcast, mask, host range, and host count details.`text_diff` for a line-by-line comparison.`generate_uuid` for one or more UUID v4 values.`jwt_validate` for structural and expiration checks.The tool schemas use Zod validation and bounded inputs. For example, the released generator implementation limits UUID batches to 100, NanoID length to 128, password length to 256, and password batches to 50. These boundaries make the tool contract easier for a client to present and enforce. The server is a thin stdio adapter around small utility handlers. Its entry point creates an MCP server, registers eight tool categories, and connects a `StdioServerTransport` . It does not expose an HTTP listener. The package lists the official TypeScript MCP SDK, `bcryptjs` , `nanoid` , and `zod` as runtime dependencies. That design is useful when the caller is an AI assistant. The assistant can select a named operation with a schema and receive a structured text result. It is less useful when you are writing normal application code. In that case, native Node.js, Python, or Go libraries avoid MCP process startup and message overhead. If the client shows no tools, first run the pinned `npx` command directly and repeat the initialize smoke test. Then check the JSON shape, executable name, Node.js version, and whether the client was restarted. A configuration path copied from macOS will not automatically be correct on Windows. Do not send secrets to debugging utilities merely because the server is local. `jwt_decode` explicitly decodes the header and payload without verifying the cryptographic signature. `jwt_validate` checks structure, JSON, expiration, and the presence of a signature string, but it also does not verify that signature. Treat its output as inspection, not authentication. The repository's security policy says the server runs locally via stdio and does not send user data to external services. That is a useful boundary, not a blanket security guarantee. Your MCP client still receives the inputs and outputs, npm installation still has a supply-chain dependency, and a local process runs with the permissions of its user. Review client permissions and pin versions when the environment is sensitive. The Dockerfile builds on Node 22 Alpine and runs the runtime image as a non-root user. That reduces one class of container risk, but it does not make arbitrary client configuration safe or prove that every dependency is harmless. Use the project's security reporting process for vulnerabilities. No. It is an MCP interface for assistants. Direct libraries are usually the better choice inside application code. No. The documented tools are local and the package has no external API dependency for its normal operation. `jwt_validate` verify a token? No. It checks structure and expiration-related fields. Use a real JWT verification library with the correct issuer and signing keys for authentication decisions. Yes. The release README documents `ghcr.io/paladini/devutils-mcp-server` and a local Docker build. The container uses stdio, so keep interactive input enabled with `docker run -i` . DevUtils MCP Server is a practical boundary between an AI assistant and a small set of deterministic developer operations. Install the released version, verify the stdio handshake, use explicit tool names, and keep authentication, secrets, and application-critical decisions outside the inspection-only helpers. Have you found a developer utility that is safer or easier to use as an explicit MCP tool than as free-form model reasoning? This tutorial was researched and edited with AI assistance. The repository, release metadata, source files, npm metadata, configuration examples, and initialize smoke test were checked against the cited primary sources before publication.