Awesome MCP servers & Skills for your agent
An awesome, browsable list of open-source MCP servers and Claude/Cursor-style Skills — search, compare, and install with the exact config snippet for your client.
Trending this week
How DReview works
From finding a tool to running it, in three steps.
Browse by category
View all categories →Guides & tutorials
An MCP server directory is only as useful as the details it shows you before you install anything. The features that matter most are server type (local or remote), implementation language, license, per-client install instructions, how a listing was verified, and whether the project is still maintained. This guide explains what each one tells you, how to read it, and which claims a directory should never make.
AWS and Azure take genuinely different approaches to connecting an agent to the cloud: AWS Labs ships over 50 small, specialized MCP servers, while Microsoft ships one unified server covering 40-plus Azure services. This guide breaks down what each actually includes, how authentication differs, why there's no vetted GCP MCP server on DReview yet despite real demand, and where Terraform, Kubernetes, and Docker fit alongside them.
Python is one of the two languages with first-class MCP SDK support, and seven Python MCP servers are already live and verified on DReview today — covering databases, documents, and messaging. This guide compares the official Python SDK and its FastMCP high-level API against TypeScript, shows what building and testing a server actually looks like, and rounds up every Python MCP server you can install right now.
Four official payment MCP servers, four different transports and tool surfaces. What each one actually exposes, and which fits your stack.
MySQL has a dedicated MCP server. Postgres doesn't — it's served through universal database gateways instead. Here's what that actually means for setup and features.
Every listing on DReview passes through a human review before it goes live — a real repo, a real license, install configs that are actually tested, not copy-pasted from a README that might be stale.
MCP servers on GitHub come from four different places — the official repo, company org accounts, community awesome-lists, and solo maintainers — each with a different built-in trust level. A real checklist (commit recency, permissions requested, license, maintainer history) for telling a trustworthy server from one that will waste your afternoon.
Claude Desktop, Claude Code, Cursor, VS Code, and Windsurf all have native MCP support in 2026 — plus a growing list of extensions like Cline, Continue, Zed, and JetBrains's AI Assistant. Here's what's actually native vs. add-on, and how to tell the difference before you assume a tool supports it.
MCP and REST API solve different problems and aren't really competing designs — REST is a convention for building HTTP APIs a human developer integrates once; MCP is a protocol for letting an AI agent discover and call tools dynamically, often by wrapping an existing REST API underneath.
Every MCP server on DReview is tagged local or remote — here's what that actually means for setup, data access, and trust, with real examples from the directory.
A practical walkthrough for building a working MCP server with the official TypeScript SDK — from scaffolding a project to testing it in the MCP Inspector and wiring it into Claude Desktop.
Stop context-switching. Start building.
Why developers are wiring MCP servers into their agents instead of hand-rolling API integrations.
What is DReview?
DReview is a free, open directory of Model Context Protocol (MCP) servers and Claude/Cursor-style Skills. Every listing includes the exact install config for Claude Desktop, plus Cursor, VS Code, and Windsurf for tools that support them, so you can go from finding a tool to running it in one copy-paste — no digging through a GitHub README to figure out the right flags.
What is an MCP server?
An MCP server is a program that exposes a specific capability — a database connection, a browser, a chat platform — to an AI agent using the Model Context Protocol, an open standard originally introduced by Anthropic. A client like Claude Desktop or Cursor connects to it, and the agent gains access to whatever tools the server defines. Servers run one of two ways: local (stdio), as a subprocess on your own machine, or remote (SSE), hosted centrally so any client can connect over HTTPS. A Skill is a lighter-weight variant of the same idea, bundled directly with a specific client instead of running as its own process.