GitHub list of ChatGPT Plugin Directory

Best MCP servers for ChatGPT and Codex

MCP servers that connect ChatGPT developer mode, Codex and the OpenAI API to your tools and data, ranked. Every server comes with a config you can copy instead of piecing it together from a README. 192 listed.

Collection · 95 MCP servers

MCP servers for Development

The most notable MCP servers in Development.

Top MCP servers

Trending MCP servers

Browse MCP servers by category

Best MCP servers: editor's picks

  1. GitHub's own server gives an agent the parts of GitHub that aren't in your local checkout: issues, pull requests and reviews, Actions runs, security alerts and code in other repositories. It answers "why did CI fail on main?" or "which PRs are waiting on my review?". In ChatGPT itself, OpenAI's GitHub plugin is usually easier.

  2. Context7 looks up current, version-specific documentation before your coding agent writes code against Next.js, Supabase, Tailwind or a smaller library, which cuts down on renamed functions and APIs that never existed. Mention the version you're on, or add "use context7" to a prompt.

  3. Playwright MCP gives a coding agent a real browser that it reads as an accessibility tree, so it finds a button by its label without a vision model. Use it to check a feature on localhost, reproduce a bug report step by step or draft a Playwright test. It runs on your machine, so it fits Codex better than ChatGPT on the web.

  4. Chrome DevTools MCP lets an agent record a performance trace, list failed network requests, read console errors with source-mapped stack traces and run a Lighthouse audit. It turns "why is this page slow?" into something the agent can investigate, fix and check again.

  5. Supabase's server reads your schema, writes and applies migrations, runs SQL, generates TypeScript types, deploys Edge Functions and runs the security and performance advisors. It acts with your developer permissions, so Supabase says not to hand it to end users.

  6. Stripe's hosted server connects an agent to your Stripe account and to Stripe's docs: look up a customer, check why charges failed, summarize your balance, or find the right API parameters while you build.

  7. Notion's hosted server at https://mcp.notion.com/mcp searches a workspace, fetches pages, creates and updates pages and databases, and works with comments. It signs in with OAuth and is the supported option; the older self-hosted package is no longer actively maintained.

  8. Linear's hosted server finds, creates and updates issues, projects and comments. It supports OAuth or a Linear API key, and a read-only endpoint is available if you only want an agent to look.

  9. Figma's remote server gives agents design context, screenshots, variables and Code Connect mappings from Figma files, and can create and edit Figma content and FigJam diagrams. It uses OAuth; a local server in the desktop app is also available.

  10. DeepWiki, by Cognition, answers questions about any public GitHub repository from generated documentation. It's free and needs no authentication, which makes https://mcp.deepwiki.com/mcp the quickest server to test a ChatGPT or API setup with.

What is an MCP server?

The Model Context Protocol (MCP) is an open standard for connecting AI clients to tools and data. An MCP server exposes tools (actions like "create issue" or "run query"), handles authentication and can return data or an interface. Any MCP client can use it: ChatGPT, Codex, the OpenAI API and many other AI tools.

Every ChatGPT app runs on an MCP server. The difference is who sets it up: apps and plugins come reviewed from ChatGPT's Plugin Directory, while an MCP server listed here is something you connect yourself, by URL for a remote server or by command for a local one.

Servers come in two kinds:

  • Remote servers run at an HTTPS URL (Streamable HTTP, or the older SSE). They work in ChatGPT, Codex and the OpenAI API, and usually sign in with OAuth.
  • Local servers run as a command on your own machine (stdio). They work in Codex and other local clients, not directly in ChatGPT on the web.

Use Remote only above the list to see servers you can connect to ChatGPT.

How to add an MCP server to ChatGPT

  1. On a Plus, Pro, Business, Enterprise or Education account, open Settings > Security and login and turn on Developer mode. Workspace admins can turn it off.
  2. Go to chatgpt.com/plugins and select the plus button.
  3. Enter a name, a short description and the server URL, including its path (usually /mcp).
  4. Pick the auth mode the server uses (OAuth, No Authentication or Mixed) and select Create.
  5. Start a new chat, open the plus menu, choose Developer mode and select your server.

OpenAI labels developer mode as elevated risk, because servers can take write actions. Connect servers you trust, and read each action before you confirm it. Step-by-step guide: How to add a custom MCP server to ChatGPT.

How to add an MCP server to Codex

Codex supports both remote and local servers, and the CLI, the ChatGPT desktop app and the IDE extension share one config file, ~/.codex/config.toml.

bash
# a local server, started by a command
codex mcp add context7 -- npx -y @upstash/context7-mcp

# a remote server, then sign in if it uses OAuth
codex mcp add linear --url https://mcp.linear.app/mcp
codex mcp login linear

In the desktop app, use Settings > MCP servers > Add server. Every listing on this site has a copyable config for Codex, ChatGPT and the API. More in Use plugins and MCP servers in Codex.

How this list is ranked

Servers are ordered by a score that combines official status, open-source traction, how complete the listing is and how recently it was added. "Official" means the server is published by the company behind the product, or by the MCP project for its reference servers. Nobody can pay for a higher position, and sponsored placements are always labelled.

MCP servers: frequently asked questions

Does ChatGPT support MCP servers?

Yes. Every ChatGPT app runs on an MCP server, and with developer mode you can connect any remote MCP server yourself by URL, including tools that write data. Developer mode is available on Plus, Pro, Business, Enterprise and Education accounts.

Can I use a local MCP server with ChatGPT?

Not directly. ChatGPT connects over the network, so it needs a public HTTPS URL; a server that runs as a local command (stdio) won't work as is. Put it behind an HTTP endpoint, use a development tunnel such as ngrok, or use OpenAI's Secure MCP Tunnel for private servers. Codex runs local servers without any of that.

Are MCP servers free?

Many are. The MCP reference servers and most open-source servers are free to run, and DeepWiki's hosted server is free with no sign-in. Servers from companies like Stripe, Supabase or Linear are free to connect but act on your account with that service, so its plan and pricing apply. Developer mode in ChatGPT needs a paid ChatGPT plan.

What's the difference between an MCP server and a ChatGPT plugin?

A plugin is the reviewed package you install from ChatGPT's Plugin Directory: usually one or more apps, each running on an MCP server, plus skills. An MCP server is the connection on its own, which you add yourself. If a product has a plugin, start there; use its MCP server when you want it in Codex, the API or another client, or when there's no plugin. More in plugins vs apps vs MCP servers vs skills.

Can I use MCP servers with the OpenAI API?

Yes. The Responses API has an mcp tool: pass a server_label and a remote server_url, and the model can call the server's tools, with require_approval controlling which calls need sign-off and allowed_tools limiting what it can use. The Agents SDK also connects to MCP servers. See MCP in the OpenAI API.

Are MCP servers safe?

A server can only do what its tools allow, but those tools act with the access you grant, and some can write or delete. Prefer official servers, use read-only modes or scopes where they exist, and keep approval on for write actions. In ChatGPT, developer mode asks you to confirm write actions before they run.

How do I test my own MCP server?

Run it locally and connect the MCP Inspector to check its tools, then expose it over HTTPS with a tunnel and add it to ChatGPT in developer mode. Test and debug an MCP server covers both steps.

Tutorials

All tutorials
Build your first ChatGPT app with an MCP server and a UI widgetBuild a minimal ChatGPT app in TypeScript: an MCP server with one tool and an MCP Apps widget, tested in MCP Inspector and connected through developer mode.Intermediate 45 minFor buildersHow to add a custom MCP server to ChatGPT with developer modeConnect your own remote MCP server to ChatGPT using developer mode: requirements, auth options, step-by-step setup, refreshing tools and fixing common errors.Intermediate 15 minFor usersTest and debug an MCP server with MCP InspectorDebug an MCP server before you connect it to ChatGPT: MCP Inspector web UI and CLI, tunnels, logs, schema and annotation checks, and OAuth debugging.Intermediate 20 minFor buildersUse plugins and MCP servers in CodexInstall plugins in the Codex CLI with /plugins or codex plugin add, add marketplaces, enable plugins per project, and connect MCP servers through config.toml.Beginner 15 minFor builders