This page in plain text
Official MCP servers
MCP is an open standard for connecting AI assistants to the systems where your work lives: files, browsers, business tools, databases. A server is one of those connections. These come from the standard’s own project or from the company whose product they connect.
What MCP is, in three lines
- One plug for many tools. Before MCP every assistant needed its own connection to every tool. Now a tool offers one server and any assistant that speaks MCP can use it.
- A server offers actions. Each server tells the assistant what it can do, such as read a file, open a page or run a query, and the assistant asks for those by name.
- You decide what it reaches. A server works with the access you give it. Nothing is connected until you connect it.
Start here
- Filesystem (filesystem), from MCP project. Lets an assistant read and write files in the folders you name, and nowhere else. The simplest way to see what MCP does.
- Fetch (fetch), from MCP project. Fetches a web page and hands the assistant its content in a form it can read.
- Git (git), from MCP project. Read, search and work with a Git repository: history, differences and changes.
- Memory (memory), from MCP project. A small store of facts the assistant can keep between conversations.
- Sequential Thinking (sequentialthinking), from MCP project. A tool for working through a problem in ordered steps that can be revised along the way.
- Time (time), from MCP project. The current time and conversions between time zones, which a model cannot know on its own.
Browser
- Playwright (microsoft/playwright-mcp), from Microsoft. Lets an agent drive a real browser: open pages, click, fill in forms and read what is on screen.
- Chrome DevTools (ChromeDevTools/chrome-devtools-mcp), from Google Chrome team. Chrome’s developer tools for coding agents: inspect a page, read the console and network, and measure performance.
Code and docs
- GitHub (github/github-mcp-server), from GitHub. GitHub’s official server: repositories, issues, pull requests and more, from inside the assistant.
- Context7 (upstash/context7), from Upstash. Up-to-date documentation for code libraries, so the agent works from current docs and not from memory.
Apps and sites
- Notion (makenotion/notion-mcp-server), from Notion. Notion’s official server, for reading and writing the pages and databases in a workspace.
- Atlassian (atlassian/atlassian-mcp-server), from Atlassian. Atlassian’s official remote server for Jira, Confluence and its other products.
- Zapier (zapier/mcp), from Zapier. The hosted Zapier server: one connection that reaches the thousands of apps Zapier already links.
- Figma (figma/mcp-server-guide), from Figma. Figma’s guide to its MCP server, which gives a coding agent the design it is meant to build.
- WordPress MCP Adapter (WordPress/mcp-adapter), from WordPress. Lets MCP clients discover and use what a WordPress site’s plugins and themes offer. The route to follow if your site, like this one, runs on WordPress.
Data and cloud
- Supabase (supabase/mcp), from Supabase. Connect a Supabase project to an assistant: its database, tables and settings.
- MongoDB (mongodb-js/mongodb-mcp-server), from MongoDB. Connect to MongoDB databases and Atlas clusters.
- MCP Toolbox for Databases (googleapis/mcp-toolbox), from Google. An open source server from Google for connecting assistants to databases.
- Cloudflare (cloudflare/mcp-server-cloudflare), from Cloudflare. Cloudflare’s servers for its own products, one for each.
- AWS (awslabs/mcp), from Amazon Web Services. Open source servers for AWS services, published by AWS Labs.
- Stripe (stripe/ai), from Stripe. Stripe’s tools for building AI products on Stripe, including its MCP server. Start in test mode.
- Sentry (getsentry/toolkit), from Sentry. Sentry’s agent tooling, so an agent can read the errors your software is throwing.
Web search
- Firecrawl (firecrawl/firecrawl-mcp-server), from Firecrawl. Firecrawl’s official server for searching, scraping and crawling the web.
- Exa (exa-labs/exa-mcp-server), from Exa. Web search and crawling from Exa, a search engine built for AI.
- Brave Search (brave/brave-search-mcp-server), from Brave. Web search through Brave’s own index.
Looking for something else? The official MCP registry and GitHub’s MCP registry list many more.
How to connect one
- Pick the one system you most often copy and paste from. Start with that.
- Follow your assistant’s instructions for adding a server: the MCP introduction explains the idea, and Connect Claude Code to tools via MCP gives the steps for Claude Code.
- Give it the least access that does the job, read-only where you can, and watch the first few things it does.
Before you install. A server acts with whatever access you hand it. Connect only servers from the maker of the product or the MCP project itself, keep private and customer data behind read-only access, and remove a server you no longer use.
Every link goes to the maker’s own repository, each opened on 20 November 2026. See also AI skills and the tools we use.
Ask Another AI Guy
I'm just Another AI Guy.
You're chatting with our fast model. Create a free account for the more thorough one, and more messages.