gx talks to your agent via the Model Context Protocol. Pick the agent you use, follow the steps, then come back to your setup checklist.
1. Add the gx plugin marketplace and install the plugin:
/plugin marketplace add gxtools/plugin
/plugin install gx@gxtools2. Or, configure the MCP server manually in ~/.claude/settings.json (no plugin):
{
"mcpServers": {
"gx": {
"type": "http",
"url": "https://gx.tools/mcp"
}
}
}1. In claude.ai, go to Settings → Connectors → Add custom MCP.
2. Paste:
Server URL: https://gx.tools/mcp
Authentication: OAuth (the prompt will open a browser window)
Scopes: read, write1. Open Cursor Settings → MCP.
2. Add to your ~/.cursor/mcp.json (or the workspace equivalent):
{
"mcpServers": {
"gx": {
"url": "https://gx.tools/mcp"
}
}
}3. Reload Cursor and authorize when prompted.
ChatGPT supports remote MCP servers on Pro, Business, Enterprise, and Edu plans.
1. In ChatGPT, open Settings → Connectors and choose Add custom connector (or Developer Mode → Add MCP serverif you're on a plan without the connectors UI).
2. Paste:
Name: gx
Server URL: https://gx.tools/mcp
Authentication: OAuth
Scopes: read, write3. In a new conversation, enable the gx connector from the tools menu before asking the model to read or write your graph.
The Codex CLI (and Codex Cloud) reads MCP servers from your ~/.codex/config.toml.
1. Append:
[mcp_servers.gx]
url = "https://gx.tools/mcp"2. Run codex (or open Codex Cloud) — the first tool call will open a browser for OAuth, then subsequent runs reuse the stored token.
3. Verify with "list my gx projects". If you're driving Codex from inside another agent, you can also point that agent at gx directly using whichever section above matches it.
From your agent, ask: "list my gx projects". You should see your projects listed.
Then try: "use gx-capture to record this" followed by something on your mind.