Codex
Codex is the default client for gavana mcp install, and it connects to the hosted endpoint over streamable HTTP with browser OAuth.
One command
gavana mcp install codexCodex is also the default client name, so bare gavana mcp install does the same thing.
The command runs Codex’s own CLI for you. What it executes is:
codex mcp add gavana --url https://app.gavana.ai/mcpYou can run those directly if you would rather not install the Gavana CLI. To see the command without executing it:
gavana mcp config codexThat prints the client, transport, endpoint, and the exact command and arguments.
Then authenticate
Codex initiates OAuth on first use. A browser opens, you sign in to Gavana, and you approve a consent screen.
The consent screen asks for all eight: canvas:read, canvas:write, asset:read, element:read, element:write, image:generate, video:generate, job:manage.
No token is ever pasted into Codex or into your shell.
Resulting permission surface
29 hosted tools. Canvas writes, Element CRUD, workflows, image and video generation, and Run observation.
Three tools — run_canvas_workflow, generate_image_in_canvas, and generate_video — reach your connected provider and can incur cost. The server requires explicit current-turn approval before calling them and never auto-retries a failed paid call.
Local stdio instead
For the 57-tool surface, print the local configuration:
gavana mcp config localThat returns the command (npx), args (-y @gavana.ai/mcp@0.2.0), and a note that the server reads your active Gavana CLI profile or inherited GAVANA_BASE_URL and GAVANA_AGENT_TOKEN variables. Add GAVANA_MCP_READ_ONLY=true to the environment if you want the local process to register only non-mutating tools.
That last point is the convenient part: if you have already run gavana auth login, the local server picks up the saved profile at ~/.config/gavana/agent.json and you configure no token at all.
Register it with Codex as a stdio server running that command. Requires Node.js 20 or newer.
@gavana.ai/mcp is published on the public npm registry at version 0.2.0, so npx -y @gavana.ai/mcp@0.2.0 resolves without extra configuration. If your environment proxies npm through a private registry that does not mirror it, use a hosted endpoint instead.
Verify
List Codex’s MCP servers and confirm gavana is connected with 11, 29, or 57 tools. Then ask it to list your canvases.
Switching endpoints later
Remove and re-add rather than editing the URL, so the old OAuth grant does not linger:
codex mcp remove gavana
gavana mcp install codex