AI agents · OpenClaw · self-hosting · automation

Quick Answer

Best MCP Servers July 2026: Updated Registry Guide

Published:

The MCP Ecosystem in July 2026

Model Context Protocol (MCP) — Anthropic’s open standard for connecting AI assistants to tools and data — has matured from novelty (late 2024) to critical infrastructure by July 2026. The official registry (registry.modelcontextprotocol.io) now lists thousands of servers, backed by Anthropic, GitHub, PulseMCP, and Microsoft.

Here are the servers that actually move the needle in production July 2026.

The Top 10 MCP Servers (July 2026)

1. Playwright MCP — Browser automation

The Microsoft-maintained Playwright MCP lets AI agents drive real browsers: click, type, navigate, screenshot, extract. It’s the default for anything involving web scraping, form filling, or QA.

  • Install: npx @playwright/mcp@latest
  • Best for: Browser automation, testing, form filling, scraping
  • Client support: Claude Desktop, Cursor, VS Code, Codex CLI

2. GitHub MCP — Repository access

GitHub’s official MCP server provides read/write to issues, PRs, files, and code. It’s authenticated via GitHub personal access token or OAuth.

  • Install: Native GitHub Copilot integration or github/github-mcp-server
  • Best for: Code review, PR triage, issue management
  • Client support: All major MCP clients

3. Figma MCP — Design context

Figma’s MCP server exposes design files as structured data — components, styles, layouts. Cursor’s Figma integration uses this to convert designs to code.

  • Install: Figma Desktop app → Preferences → MCP
  • Best for: Design-to-code workflows
  • Client support: Cursor, Claude Desktop, Windsurf

4. Supabase MCP — Database access

Query and mutate Postgres, manage RLS policies, and inspect schemas — all from the AI agent. Supabase’s server is production-ready with row-level security.

  • Install: npx @supabase/mcp-server
  • Best for: Full-stack apps with Supabase backends
  • Client support: All major MCP clients

5. Notion MCP — Docs and databases

Notion’s official MCP server (launched Q1 2026) exposes pages, databases, and comments. Excellent for knowledge-base RAG without embeddings.

  • Install: Notion → Settings → Integrations → MCP
  • Best for: Knowledge management, internal docs, project tracking
  • Client support: Claude Desktop, Cursor

6. Slack MCP — Team chat

Anthropic’s Slack MCP server reads channels, DMs, and threads. Excellent for meeting-note synthesis and team-context grounding.

  • Install: mcp-server-slack (Anthropic official)
  • Best for: Team-context grounding, meeting synthesis
  • Client support: Claude Desktop, Claude Code

7. Filesystem MCP — Local files

The reference filesystem server (bundled with the official SDK) is the simplest way to give an AI agent read/write access to a local directory. Use with caution — set narrow paths.

  • Install: Built into @modelcontextprotocol/sdk
  • Best for: Local codebases, local docs
  • Client support: All major MCP clients

8. Jira / Atlassian MCP — Project management

Atlassian’s official MCP server (Jira, Confluence) exposes tickets, epics, and wiki pages. Enterprise-grade auth.

  • Install: atlassian/mcp-server
  • Best for: Enterprise project tracking, requirements gathering
  • Client support: All major MCP clients

9. Context7 MCP — Fresh library docs

Context7 pulls the latest library documentation (React, Next.js, Prisma, etc.) into the AI’s context — solves the “AI knows old API” problem.

  • Install: npx -y @upstash/context7-mcp
  • Best for: Coding with fast-moving libraries
  • Client support: Cursor, Claude Code, all major clients

10. Serena MCP — Code intelligence

Serena provides semantic code navigation (find definitions, references, callers) across the codebase — LSP-quality context for AI agents.

  • Install: pip install serena-mcp
  • Best for: Large codebases where LLM context doesn’t fit
  • Client support: Claude Desktop, Cursor, VS Code

The Registries

RegistryWhat it isBest for
Official RegistryCanonical metadata (Anthropic + GitHub + Microsoft + PulseMCP)Verified server metadata
Smithery7,000+ installable servers, hosted + localDiscovery + one-command install
GlamaMetaregistry aggregating multiple sourcesBroad discovery
MCP.soSearchable directoryQuick lookup
TrueFoundry MCP GatewayEnterprise governance layerAccess control + audit logging

How to Install (Fastest Path)

Claude Desktop (macOS example):

# Install via Smithery
npx @smithery/cli install playwright-mcp --client claude

# Or manually edit:
open ~/Library/Application\ Support/Claude/claude_desktop_config.json

Cursor:

# Cursor Settings → MCP → Add Server
# Or edit:
open ~/.cursor/mcp.json

Codex CLI:

codex config mcp add <server-name>

What Changed in Q2 2026

  • Official registry launched (Q1 2026, matured Q2) — canonical server metadata
  • MCP 1.4 released — better auth, streaming, and error handling
  • OpenAI Codex CLI added MCP support — bringing MCP into the OpenAI ecosystem
  • Enterprise gateways emerged (TrueFoundry, Glama Enterprise) — access control, audit
  • Smithery hit 7,000 servers — up from ~1,500 at start of 2026

The Bottom Line

If you’re setting up an MCP-enabled AI workflow in July 2026, the minimum-viable stack is:

  1. Playwright MCP — for anything web
  2. GitHub MCP — for anything code
  3. Filesystem MCP — for anything local
  4. Context7 MCP — for anything with fast-moving library docs

Add domain-specific servers (Figma, Supabase, Notion, Slack) as your workflow demands. Install via Smithery for the fastest experience; edit configs manually for production stacks with tight access control.

Sources