Awesome MCP Servers List (70+) — Live-Tested Catalog for 2026

This is a curated list of the best MCP (Model Context Protocol) servers for AI development in 2026. Each server has been selected based on active maintenance, documentation quality, and community adoption. Updated July 2026 — the hyperscalers and large platforms have now arrived: AWS made its managed MCP server generally available on May 6, X shipped two hosted servers on June 30, and Snowflake now offers a warehouse-managed endpoint. The list is 69 servers across 11 categories.

Remote HTTP Servers — Connect with Just a URL

These servers offer hosted HTTP endpoints you can connect to without any local installation. The remote MCP ecosystem has grown steadily through 2026 — from 16 servers in January to 25+ by April to 30 in July, as Atlassian, HubSpot, Linear, Slack, Sentry, Neon and Vercel launched remote endpoints, and AWS, X and Snowflake followed with managed ones.

The new arrivals — and the auth catch on each

Three of the servers above landed after this list's April update, and each one authenticates differently from the OAuth-and-a-URL pattern the rest of the ecosystem settled on. That matters more than the tool counts, because it is where setup actually fails.

AWS MCP Server (GA May 6, 2026) is the most capable server on this list and the most awkward to connect. A single call_aws tool fronts more than 15,000 AWS API operations, run_script executes Python server-side in a sandbox with your IAM permissions but no network access, and search_documentation / read_documentation work with no authentication at all. The catch: AWS authenticates with IAM SigV4, and MCP only specifies OAuth 2.1. The two don't meet, so AWS ships an open-source MCP Proxy for AWS to bridge them — you cannot simply paste the URL into a client and expect it to connect. Requests are logged to CloudTrail and metered in CloudWatch under the AWS-MCP namespace, which makes it one of the few servers here that is genuinely auditable in production.

X (June 30, 2026) ships two servers with opposite requirements. The API server at api.x.com/mcp runs Streamable HTTP and expects OAuth 2.0 PKCE through X's xurl mcp local bridge, which refreshes tokens so the model inherits your account's scopes; an app-only bearer token also works but gives you read-only access with no user context. Note that early coverage described the server as read-only across the board — X's own docs list bookmark management and article publishing as supported writes, subject to tighter rate limits than reads. The docs server at docs.x.com/mcp needs no auth at all and exposes just two tools, search_x and get_page_x. It is the single easiest server in this entire list to try.

Snowflake's managed server has no shared URL — the endpoint lives inside your account at /api/v2/databases/{database}/schemas/{schema}/mcp-servers/{name}, so the server is a Snowflake object you provision rather than a address you look up. It exposes up to 50 tools spanning Cortex Analyst (natural language to SQL), Cortex Search, Cortex Agents, and direct SQL execution that can be pinned read-only. OAuth 2.0 is the recommended auth; programmatic access tokens work but Snowflake's own docs warn about token leakage.

The broader pattern across all three: the platforms with the most to lose from a runaway agent are the ones bolting their existing authorization model onto MCP rather than adopting plain OAuth. Expect more per-account URLs and more bridges, not fewer.

Database Servers

Developer Tools

AI Tools & Memory

These servers extend AI capabilities with persistent memory, advanced reasoning, and real-time search — essential for building agents that remember context across sessions.

Monitoring & Observability

Track errors, query metrics, and manage incidents — these servers connect AI to your monitoring stack.

Productivity & Project Management

Payments & E-Commerce

Cloud & Infrastructure

Automation & Workflows ✨ New

Connect AI to your automation pipelines. These servers let Claude trigger workflows, pass data between apps, and orchestrate multi-step processes without code.

Design & Creative

SEO & Analytics

Search & Data

Quick Setup Example

Add remote MCP servers to your Claude Desktop, Cursor, or Claude Code config. The HTTP servers listed above require no local installation — just add the URL.

Claude Desktop / Cursor

{
 "mcpServers": {
 "supabase": {
 "url": "https://mcp.supabase.com/mcp"
 },
 "github": {
 "type": "http",
 "url": "https://api.githubcopilot.com/mcp/"
 },
 "context7": {
 "type": "http",
 "url": "https://mcp.context7.com/mcp"
 },
 "memory": {
 "command": "npx",
 "args": ["-y", "@modelcontextprotocol/server-memory"]
 },
 "sequential-thinking": {
 "command": "npx",
 "args": ["-y", "@modelcontextprotocol/server-sequential-thinking"]
 }
 }
}

Claude Code (~/.claude.json)

{
 "mcpServers": {
 "supabase": {
 "url": "https://mcp.supabase.com/mcp"
 },
 "github": {
 "url": "https://api.githubcopilot.com/mcp/"
 },
 "context7": {
 "url": "https://mcp.context7.com/mcp"
 },
 "memory": {
 "command": "npx",
 "args": ["-y", "@modelcontextprotocol/server-memory"]
 }
 }
}

Getting Started

• Claude Desktop Setup Guide — Configure MCP servers in Claude
• Cursor IDE Setup Guide — Configure MCP servers in Cursor
• MCP Playground — Test any server instantly online
• MCP Registry — Browse 10000+ MCP servers