MCP server directory
The Model Context Protocol is how an AI app plugs into the outside world: files, databases, APIs, and tools, through one standard interface instead of a custom integration each time. Here are the servers worth knowing, grouped by what they actually do.
Code & developer
- Filesystem
Read, write, and search local files in a sandboxed root.
- Git / GitHub
Browse repos, read and open PRs, manage issues and branches.
- Puppeteer / Playwright
Drive a headless browser to scrape, test, or automate the web.
- Sentry
Pull error events and issues into the model for triage.
Data & databases
- Postgres
Query a database read-only with schema introspection.
- SQLite
Local database access for analysis and prototyping.
- Supabase
Tables, SQL, and edge functions for a Postgres backend.
Productivity & knowledge
- Notion
Read and write pages, databases, and comments.
- Slack
Read channels and threads, send and schedule messages.
- Google Drive
Search and read documents across a Drive.
Web, search & cloud
- Brave / web search
Live web search results the model can read and cite.
- Fetch
Pull a URL and convert it to clean text for the model.
- Cloud providers
Deploy and inspect infra (e.g. Vercel, Netlify, AWS) via tools.
What MCP actually changes
Before MCP, every tool an AI could use was a bespoke integration glued to one app. MCP turns that into a protocol: a server exposes tools, resources, and prompts, and any compatible client can use them. Write the connector once, and every MCP-aware agent can pick it up. That is why the ecosystem grew so fast, and why "which MCP servers should I run" is now a real question.
If you are new to it, start with the deeper write-ups in the MCP guide and where MCP fits in the agent stack. Then wire one server into an agent loop, which is exactly the pattern in loop engineering.
This directory is curated and grows over time. For the full, current list and install instructions, the official servers repository above is the source of truth.