Ecosystem
Awesome MCP Servers: curated by category (May 2026)
- Reference
- Tooling
MCP servers worth connecting to Claude Code. Curated by category: data, design, dev, browser, security, productivity. GitHub data verified 2026-05-11.
TL;DR
- MCP (Model Context Protocol) is the canonical way to plug external tools and data into Claude Code. In 2026, the open-source standard moved from experimental to dominant.
- Aggregated references:
punkpeye/awesome-mcp-servers(ouvre un nouvel onglet) (86k★) andmodelcontextprotocol/servers(ouvre un nouvel onglet) (85k★, official repo). - 6 categories cover 80% of needs: dev (GitHub, Chrome DevTools, Playwright, Serena), data (Supabase), design (shadcn, Figma), web nav (Context7), automation (n8n, activepieces), builder frameworks (FastMCP).
- Advice: start with 3 servers tied to your stack, not 30 at once. MCPs consume context and can slow Claude down.
1. Awesome lists & official repos
Keep these as permanent references.
| Repo | ⭐ | What's special |
|---|---|---|
punkpeye/awesome-mcp-servers (ouvre un nouvel onglet) | 86,708 | THE reference awesome list for MCP servers. Sorted by category, very actively updated. |
modelcontextprotocol/servers (ouvre un nouvel onglet) | 85,460 | Official protocol repo: reference servers (filesystem, postgres, memory, brave-search…). |
2. Dev tools (must-haves)
The MCPs that live in developer terminals.
| Repo | ⭐ | What it plugs in |
|---|---|---|
ChromeDevTools/chrome-devtools-mcp (ouvre un nouvel onglet) | 39,234 | Chrome DevTools for coding agents. Inspect DOM, capture network, run JS, debug live. |
microsoft/playwright-mcp (ouvre un nouvel onglet) | 32,364 | Playwright MCP official Microsoft. Browser automation, E2E tests, screenshots, scraping. |
github/github-mcp-server (ouvre un nouvel onglet) | 29,724 | Official GitHub MCP. Issues, PRs, search code, releases — straight from the agent. |
oraios/serena (ouvre un nouvel onglet) | 24,086 | Serena: semantic retrieval + editing for agents. "The IDE for your agent". |
3. Data & docs
When Claude needs to read or write to external stores.
| Repo | ⭐ | Plugs into |
|---|---|---|
upstash/context7 (ouvre un nouvel onglet) | 55,037 | Context7: up-to-date library docs for LLMs. Prevents version and API hallucinations. |
supabase-community/supabase-mcp (ouvre un nouvel onglet) | 2,680 | Supabase: connect Postgres + auth + edge functions + storage to the AI assistant. |
4. Frameworks to build your own MCP
When public servers aren't enough and you need to create your own.
| Repo | ⭐ | Stack |
|---|---|---|
PrefectHQ/fastmcp (ouvre un nouvel onglet) | 25,116 | FastMCP Python — "the fast, Pythonic way to build MCP servers and clients". The most popular. |
tadata-org/fastapi_mcp (ouvre un nouvel onglet) | 11,857 | FastAPI MCP: turns an existing FastAPI app into an MCP server. Zero-effort bridge. |
punkpeye/fastmcp (ouvre un nouvel onglet) | 3,108 | FastMCP TypeScript: TS equivalent of the Python framework. |
5. Automation & workflow (multi-MCP)
To orchestrate multiple MCPs together, trigger on events.
| Repo | ⭐ | What's special |
|---|---|---|
n8n-io/n8n (ouvre un nouvel onglet) | 187,464 | n8n: no-code/low-code workflow automation platform with native AI capabilities. Combines MCP + 400+ integrations. |
activepieces/activepieces (ouvre un nouvel onglet) | 22,145 | Activepieces: ~400 MCP servers for AI agents, AI automation + workflows. Open-source Zapier alternative. |
mcp-use/mcp-use (ouvre un nouvel onglet) | 9,928 | mcp-use: abstraction layer to use multiple MCPs in LangChain pipelines. |
6. Security, scraping & content
For Claude when it has to fetch data from outside.
| Repo | ⭐ | For who |
|---|---|---|
D4Vinci/Scrapling (ouvre un nouvel onglet) | 48,845 | Scrapling: adaptive web scraping framework (single request → full crawl). MCP compatible. |
assafelovic/gpt-researcher (ouvre un nouvel onglet) | 26,989 | GPT Researcher: autonomous agent that does deep multi-source research. MCP-compatible via wrapper. |
mrexodia/ida-pro-mcp (ouvre un nouvel onglet) | 8,420 | IDA Pro MCP: reverse engineering assistant for binaries. Niche but excellent. |
7. Learning to use MCPs
Not an MCP itself, but essential for getting started.
| Repo | ⭐ | What |
|---|---|---|
microsoft/mcp-for-beginners (ouvre un nouvel onglet) | 16,073 | Official Microsoft course: "MCP for beginners". Curriculum, exercises, hands-on. |
How to choose
Three principles:
- Don't plug everything: each active MCP consumes context at startup. Limit yourself to 5-7 simultaneous servers, otherwise Claude slows down.
- Profiles by context: use MCP profiles to have a "dev" setup different from "SEO audit" or "security audit" (cf.
/advanced/mcp-profiles). - Trust the official ones: for production, prefer vendor MCPs (GitHub, Microsoft, Supabase, Anthropic). Community MCPs are great to experiment.
Next steps
- Understand the MCP architecture:
/mcp/what-are-mcps. - Install your first MCP:
/mcp/setup. - MCP profiles by work context:
/advanced/mcp-profiles. - Create your own MCP (TypeScript):
/mcp/create-mcp-typescript. - Create your own MCP (Python):
/mcp/create-mcp-python.