best of · updated September 2026

The 15 best open-source MCP servers

The 15 open-source MCP servers worth your time, out of 4,077 licensed ones we track. Ranked on monthly installs, GitHub stars, and commit recency. Checked nightly.

Every server here carries a real open-source license, so you can read the code before you trust it with your data, fork it when you need a private variant, and pin the exact version you audited. License terms differ: the page for each server names its license, and the table below shows which licenses dominate.

01 Playwright MCP ★ 37,231 · commit 2026-09-17 A Model Context Protocol (MCP) server that provides browser automation capabilities using Playwright. Apache-2.0 · 23,360,601 installs/mo 02 Chrome DevTools MCP ★ 52,170 · commit 2026-09-17 Chrome DevTools for agents (chrome-devtools-mcp) lets your coding agent (such as Antigravity, Claude, Cursor or Copilot) control and inspect a live Chrome browser. Apache-2.0 · 9,332,843 installs/mo 03 Context7 ★ 62,152 · commit 2026-09-18 Up-to-date code docs for any prompt MIT · 3,719,962 installs/mo 04 Firebase MCP ★ 4,470 · commit 2026-09-18 The Firebase Command Line Interface (CLI) Tools can be used to test, manage, and deploy your Firebase project from the command line. MIT · 10,177,335 installs/mo 05 SeleniumBase MCP ★ 13,024 · commit 2026-09-18 All-in-one Browser Automation Framework: Web Crawling / Testing / Scraping / Stealth. MIT · 2,076,176 installs/mo 06 Figma Context MCP (glips) ★ 15,874 · commit 2026-09-18 Give your coding agent access to your Figma data. MIT · 307,488 installs/mo 07 Azure MCP Server ★ 3,681 · commit 2026-09-17 All Azure MCP tools to create a seamless connection between AI agents and Azure services. MIT · 524,104 installs/mo 08 Edgar Tools ★ 2,726 · commit 2026-09-17 EdgarTools is a Python library for accessing SEC EDGAR filings as structured data. MIT · 850,518 installs/mo 09 Agent Device (callstackincubator) ★ 4,641 · commit 2026-09-18 Mobile app automation and verification for AI coding agents. MIT · 530,730 installs/mo 10 Supabase ★ 2,910 · commit 2026-09-17 The Model Context Protocol (MCP) standardizes how Large Language Models (LLMs) talk to external services like Supabase. Apache-2.0 · 462,519 installs/mo 11 Claude Flow ★ 72,753 · commit 2026-09-18 An agent meta-harness for Claude Code and Codex. MIT · 77,159 installs/mo 12 Desktop Commander ★ 9,633 · commit 2026-09-17 Work with code and text, run processes, and automate tasks, going far beyond other AI editors - while using host client subscriptions instead of API token costs. MIT · 278,322 installs/mo 13 Xcode Build MCP ★ 6,397 · commit 2026-09-11 A Model Context Protocol (MCP) server and CLI that provides tools for agent use when working on iOS and macOS projects. MIT · 530,072 installs/mo 14 Ha MCP ★ 4,770 · commit 2026-09-18 A comprehensive Model Context Protocol (MCP) server that enables AI assistants to interact with Home Assistant. MIT · 344,083 installs/mo 15 GitLab MCP ★ 1,984 · commit 2026-09-16 📖 Documentation → Setup guides, environment variables, and the full tool reference live on the hosted docs site. MIT · 497,724 installs/mo

Which licenses dominate

Across all 4,077 licensed servers we track:

MIT 3,031 servers
Apache-2.0 764 servers
AGPL-3.0 165 servers
GPL-3.0 51 servers
ISC 12 servers
MPL-2.0 10 servers

Quick answers

Which license do most open-source MCP servers use?

MIT, by a wide margin: 3,031 of the 4,077 licensed servers we track use it, followed by Apache-2.0 (764). 3,823 carry permissive licenses overall.

Can I use these MCP servers in a commercial product?

Most carry permissive licenses (MIT, Apache-2.0, BSD) that generally allow commercial use with attribution. Copyleft licenses (GPL, AGPL) add conditions. Each server's page names its exact license; check it before you ship.

How this list is ranked

Ranked from live registry + GitHub data: monthly installs, GitHub stars, and commit recency, recomputed nightly. A server with no declared license, or only a NOASSERTION stub, is not listed here regardless of popularity. No pay-to-rank, ever. Full recipe: how we verify.