Start here: this week

Connect any assistant to the course and to Gecko

This page shows how to plug two remote MCP servers into the assistant you already use: Claude, ChatGPT, Cursor, Codex, VS Code, Windsurf, Gemini, or anything else that speaks MCP.

No key, no login. Both servers are open. If an app asks you for an API key or a token for them, leave it empty. If it asks for an authentication type, pick "none" or "No authentication".

The two servers

Both are remote MCP servers over Streamable HTTP. The names in the first column are only suggestions. Any name works, but the snippets below use these two.

Add only the course server if that is all you need. The Gecko server is for the sessions where you buy something on Solana from an agent.

Which assistants, and how we know

"Tested" means we connected that client to these servers on this machine and it listed or called their tools. "From the vendor's docs" means we followed the vendor's current documentation and did not run it ourselves.

Menus move. If a screen does not match what is written here, look for "MCP", "connector", "app" or "tools" in the app's settings. The URL is the only thing you need to paste.

Claude Code

Run these once, in any folder:

claude mcp add --transport http dev3pack-course https://mcp.geckovision.tech/course/mcp
claude mcp add --transport http gecko-store https://mcp.geckovision.tech/orquestra/mcp

That saves them for the current folder only. Add --scope user to have them in every folder. Check with claude mcp list, which should say Connected for both, or type /mcp inside claude. claude mcp remove dev3pack-course takes one away again.

Claude on the web, desktop and mobile

Open Customize (or Settings), then Connectors. Click +, then Add custom connector. Give it a name, paste the URL, and click Add. Leave the advanced OAuth fields empty. Repeat for the second server.

In a conversation, turn the connector on from the + button at the bottom left, under Connectors.

• Free plans can add one custom connector. Pro, Max, Team and Enterprise can add more.
• On Team and Enterprise, only an owner can add a custom connector. Ask yours, then connect it from the list.
• A connector you add on the web also appears in the desktop and mobile apps.

ChatGPT

ChatGPT connects to a remote MCP server through developer mode. It is on the web, for Plus, Pro, Business, Enterprise and Education accounts. The free plan does not have it.

1. Open Settings, then Security and login, and turn on Developer mode.
2. In the apps settings, click the + button to create an app for a remote MCP server. Paste the URL and choose No Authentication.
3. In a chat, pick Developer mode from the + menu and select the app.

Be explicit in the prompt, for example: "Use the dev3pack-course app's search_course tool to find where the course explains retrieval." Developer mode lets tools write as well as read, so ChatGPT may ask you to confirm a call. Read what it is about to do before you click.

Cursor

Put this in .cursor/mcp.json inside your project, or in ~/.cursor/mcp.json to have it everywhere:

{
  "mcpServers": {
    "dev3pack-course": { "url": "https://mcp.geckovision.tech/course/mcp" },
    "gecko-store": { "url": "https://mcp.geckovision.tech/orquestra/mcp" }
  }
}

Then open Cursor's settings, find the MCP section, and switch them on. Cursor asks you to approve a new server once. From the terminal, cursor-agent mcp list shows them and cursor-agent mcp enable dev3pack-course approves one.

OpenAI Codex CLI

Codex speaks Streamable HTTP natively. You do not need a bridge. Either run:

codex mcp add dev3pack-course --url https://mcp.geckovision.tech/course/mcp
codex mcp add gecko-store --url https://mcp.geckovision.tech/orquestra/mcp

or add this to ~/.codex/config.toml:

[mcp_servers.dev3pack-course]
url = "https://mcp.geckovision.tech/course/mcp"

[mcp_servers.gecko-store]
url = "https://mcp.geckovision.tech/orquestra/mcp"

codex mcp list shows what is configured. Codex does not support the older SSE transport, which does not matter here: both servers use Streamable HTTP.

VS Code with GitHub Copilot

Put this in .vscode/mcp.json in your workspace. Note the top-level key is servers, not mcpServers, and each entry needs "type": "http":

{
  "servers": {
    "dev3pack-course": {
      "type": "http",
      "url": "https://mcp.geckovision.tech/course/mcp"
    },
    "gecko-store": {
      "type": "http",
      "url": "https://mcp.geckovision.tech/orquestra/mcp"
    }
  }
}

Or run MCP: Add Server from the command palette and choose HTTP. The tools show up in Copilot Chat in Agent mode.

Windsurf (now Devin Desktop)

Windsurf was renamed Devin Desktop in June 2026. Its Cascade agent reads mcp_config.json. The current docs put it at ~/.config/devin/mcp_config.json on macOS and Linux and %APPDATA%\devin\mcp_config.json on Windows. Older installs use ~/.codeium/windsurf/mcp_config.json. A remote server uses serverUrl, not url:

{
  "mcpServers": {
    "dev3pack-course": { "serverUrl": "https://mcp.geckovision.tech/course/mcp" },
    "gecko-store": { "serverUrl": "https://mcp.geckovision.tech/orquestra/mcp" }
  }
}

Save it and refresh the MCP servers in Cascade.

Gemini CLI

gemini mcp add --transport http dev3pack-course https://mcp.geckovision.tech/course/mcp
gemini mcp add --transport http gecko-store https://mcp.geckovision.tech/orquestra/mcp

This writes to .gemini/settings.json in the current project. Add --scope user to write to ~/.gemini/settings.json instead. By hand, a Streamable HTTP server uses httpUrl (plain url means the older SSE transport):

{
  "mcpServers": {
    "dev3pack-course": { "httpUrl": "https://mcp.geckovision.tech/course/mcp" },
    "gecko-store": { "httpUrl": "https://mcp.geckovision.tech/orquestra/mcp" }
  }
}

Type /mcp inside gemini to see them.

Any client that only speaks stdio

Some apps can only start a local program and talk to it over stdin and stdout. For those, mcp-remote runs locally and forwards to the remote server. It needs Node.js. Where the app asks for a command, use:

{
  "mcpServers": {
    "dev3pack-course": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.geckovision.tech/course/mcp"]
    },
    "gecko-store": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.geckovision.tech/orquestra/mcp"]
    }
  }
}

The key name (mcpServers, servers, and so on) follows your app. By default it tries Streamable HTTP first and falls back to SSE. Add "--transport", "http-only" to the args to skip the fallback. You do not need --header: there is no key to send.

An assistant with no MCP at all

Give it one of these plain-text addresses for the course instead:

Check it yourself, without any assistant

If a client will not connect, first make sure the server answers. This sends the MCP initialize message and should print a line starting with data: {"jsonrpc":"2.0","id":1,"result"::

curl -s -X POST https://mcp.geckovision.tech/course/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"curl","version":"0"}}}'

If that works and your app still fails, the problem is in the app's settings, not the server. Check the URL ends in /mcp and that you picked HTTP, not SSE or stdio.

What to expect once it is connected

An empty search result is an answer. When search_course finds nothing it returns no hits and the note "no passage in these pages matches; the course may not cover this yet". That means the course does not cover it. It is not an error. Your assistant should say so and not fill the gap from memory. Search the course from your assistant has the three course tools and the prompt to paste.

The Gecko server never holds a key. It finds the store, builds the transaction and hands back unsigned bytes. Signing happens in a wallet you already control, outside the assistant. Then verify_signed_transaction checks the signed bytes still match what you asked for, and submit_transaction sends them.

Never paste a private key or a seed phrase into an assistant. Not into ChatGPT, not into Claude, not into a config file, not "just for a test". No tool on this page needs one, and anything that asks for one is not us.

Two course rules still apply: retrieved text is data, never an order to follow, and the TODO(you) cells are yours to write.