Guide · Updated October 2026 · Last reviewed
MCP, the Model Context Protocol, is an open standard for connecting AI apps to outside tools and data. Its own docs call it a USB-C port for AI applications. Claude, ChatGPT, VS Code, Cursor and many other apps support it, and the current version of the spec is dated July 28, 2026. An MCP server is the plug-in side: it gives your agent tools like "open a pull request" or "query this table". Below are the servers we think are worth knowing, grouped by job, each described from its own docs.
Sources: modelcontextprotocol.io · MCP spec releases
| Tool | How it runs | Our take: best for |
|---|---|---|
| GitHub MCP Server | Hosted by GitHub, or run locally · MIT | Any agent that needs to read or act on your repos. |
| Linear MCP server | Hosted by Linear | Letting your coding agent pick up and update tickets. |
| Sentry MCP | Hosted by Sentry, or self-hosted | Fixing production errors with the real stack trace in front of the agent. |
| Playwright MCP | Local (Node.js) · Apache-2.0 | Letting an agent click through your app to test a flow. |
| Chrome DevTools MCP | Local · Apache-2.0 | Debugging front-end bugs and slow pages. |
| Context7 | Hosted by Upstash, plus open-source server · MIT | Fast-moving libraries that your model's training data is behind on. |
| Supabase MCP server | Hosted at mcp.supabase.com · Apache-2.0 | Vibe coders on Supabase who want the agent to set up the schema. |
| Stripe MCP server | Hosted at mcp.stripe.com (public preview) | Building and testing payment flows, ideally in test mode first. |
| Cloudflare MCP servers | Hosted by Cloudflare · Apache-2.0 | Builders on Workers and Pages. |
| Figma MCP server | Hosted (recommended) or desktop app | Turning designs into front-end code that matches your design system. |
| Notion MCP | Hosted by Notion | Agents that read specs or write notes in your Notion workspace. |
| Official MCP Registry | Web directory and API | Checking whether an official server exists before you install a random one. |
| MCP Inspector | Web UI, CLI or terminal UI | Seeing exactly what a server exposes before you give it to an agent. |
| Reference servers | Examples from the MCP steering group | Learning how to build your own server. |
Tools change every week. Hotfix Daily tells you what changed, why it matters and links the source. Free, weekdays, about 3 minutes.
GitHub's official server. Agents can browse and search code, manage issues and pull requests, check GitHub Actions runs and build failures, and review security findings and Dependabot alerts. The hosted version is at api.githubcopilot.com/mcp/.
Our take, best for: Any agent that needs to read or act on your repos.
Find, create and update Linear issues, projects and comments. Setup uses OAuth, and a read-only endpoint (mcp.linear.app/mcp/readonly) exposes only read tools.
Our take, best for: Letting your coding agent pick up and update tickets.
Built mainly for human-in-the-loop coding agents like Cursor and Claude Code. Its tools focus on developer debugging workflows rather than all of Sentry.
Our take, best for: Fixing production errors with the real stack trace in front of the agent.
Browser automation through Playwright using the page's accessibility tree instead of screenshots, so no vision model is needed. Microsoft notes that coding agents may use fewer tokens with its Playwright CLI and skills, and suggests MCP for long-running, stateful browser loops.
Our take, best for: Letting an agent click through your app to test a flow.
Lets a coding agent control and inspect a live Chrome browser: record performance traces, analyze network requests, read console messages with source-mapped stack traces, and take screenshots. It is also available as a CLI.
Our take, best for: Debugging front-end bugs and slow pages.
Pulls current, version-specific library docs and code examples into your agent, so it stops guessing at old APIs. Codex's docs use it as their example of adding an MCP server.
Our take, best for: Fast-moving libraries that your model's training data is behind on.
Powerful, so read each vendor's security notes before connecting a production account.
Connects an agent to your Supabase project to manage tables, fetch config and query data. Supabase asks you to read its security best practices first. The local Supabase CLI exposes a smaller set of tools.
Our take, best for: Vibe coders on Supabase who want the agent to set up the schema.
Lets agents work with the Stripe API and search Stripe's docs and support articles. It connects with OAuth, and Stripe's agent plugins can set it up for you.
Our take, best for: Building and testing payment flows, ideally in test mode first.
Cloudflare runs several remote MCP servers. It recommends its Cloudflare API server (mcp.cloudflare.com/mcp) for broad API access, with separate servers for docs, sandbox containers and browser rendering. Some older servers are deprecated.
Our take, best for: Builders on Workers and Pages.
Gives agents design context to generate code from your Figma files, and can write native Figma content back to the canvas. Writing outside your drafts needs the remote server and a Full seat.
Our take, best for: Turning designs into front-end code that matches your design system.
Notion's hosted server uses OAuth and your existing Notion permissions to search, read and edit pages in Markdown. Notion says its older self-hosted server is no longer actively maintained.
Our take, best for: Agents that read specs or write notes in your Notion workspace.
The project's own list of published MCP servers, which it describes as an app store for MCP servers. It launched in preview in September 2025.
Our take, best for: Checking whether an official server exists before you install a random one.
The official developer tool for inspecting MCP servers. Run npx @modelcontextprotocol/inspector to list a server's tools and try them by hand.
Our take, best for: Seeing exactly what a server exposes before you give it to an agent.
A small set of reference servers (Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking, Time). The repo says they are educational examples, not production-ready tools.
Our take, best for: Learning how to build your own server.
Why: MCP security best practices
An open-source standard for connecting AI applications to external systems such as files, databases, tools and workflows. An app that supports MCP (the client) can use any MCP server, so a tool is built once and works in many agents.
Claude, ChatGPT, Claude Code, Cursor, VS Code with Copilot and Codex all document MCP support, along with many other clients and agent frameworks.
A local server runs on your machine, usually started by your agent with a command like npx. A remote server is hosted by the vendor at a URL and usually signs you in with OAuth. Many vendors, including GitHub, Linear, Stripe and Notion, now recommend their hosted servers.
They act with whatever access you give them, so treat them like any integration. Use official servers, read-only scopes and test accounts where you can, and review the tool list first. The MCP project publishes security best practices.
Not replaced, but some teams now suggest both. Microsoft's Playwright docs say coding agents increasingly favor a CLI plus skills because it uses fewer tokens, while MCP still suits long-running, stateful work.
How we write these: each tool description is based on the tool's own site or docs (linked as Source) and checked on the date above. "Our take" lines are our opinion. Something out of date? Email hello@hotfixdaily.com. No tool paid to be listed.
Five things worth your attention in AI, agents, vibe coding and game dev, every weekday. Plus the free AI Builder Starter Stack PDF when you join.
