VS Code
VS Code is a standard MCP client, so Gavana works with it — but this is the one supported client the Gavana CLI has no helper for.
gavana mcp install supports codex and claude. gavana mcp config supports cursor, chatgpt, and local. VS Code is not in either list, and passing vscode as a client name is a usage error. Configure it by hand using the values below.
The Gavana side
These are the facts that matter, and they are identical for every client:
| Hosted MCP | |
|---|---|
| URL | https://app.gavana.ai/mcp |
| Transport | Streamable HTTP |
| Auth | Browser OAuth, PKCE, no token to paste |
| Tools | 29 |
VS Code discovers everything else — authorization endpoint, token endpoint, PKCE requirement, scopes — from Gavana’s published OAuth metadata. There is nothing to configure beyond the URL.
Configure it
VS Code reads MCP servers from an mcp.json file, with workspace-level and user-level variants. The exact file location and schema belong to VS Code and change between releases, so confirm the shape against VS Code’s own MCP documentation for your version. A remote HTTP server entry looks like this:
{
"servers": {
"gavana": {
"type": "http",
"url": "https://app.gavana.ai/mcp"
}
}
}VS Code also offers Add MCP Server from the command palette, which writes the same entry for you. Choose the HTTP or remote server option and paste the URL — that route avoids depending on the file schema entirely, and is the one to prefer.
Either way, VS Code opens a browser for Gavana sign-in and consent the first time the server is used.
Resulting permission surface
29 hosted tools. The consent screen asks for all eight scopes. Three of these tools — run_canvas_workflow, generate_image_in_canvas, generate_video — can charge your connected provider.
An agent inside your editor can create canvas nodes and start paid generations. Start with Inspect before you write, and require explicit current-turn approval before any paid call.
Local stdio instead
If you need the 57-tool local surface, VS Code can run it as a stdio server. You need Node.js 20 or newer and an Agent Access token. The command and environment Gavana expects are:
command: npx
args: -y @gavana.ai/mcp@0.2.0
env: GAVANA_BASE_URL=https://app.gavana.ai
GAVANA_AGENT_TOKEN=<your Agent Access token>Express that in whatever shape your VS Code version’s mcp.json uses for stdio servers. VS Code supports input prompts for secrets — use one for the token rather than writing it into a workspace file that could be committed.
Optional narrowing, without changing the token:
GAVANA_MCP_READ_ONLY=true
GAVANA_MCP_TOOLSETS=canvas,assets
GAVANA_MCP_EXCLUDE_TOOLS=node_delete,job_cancel@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
Open VS Code’s MCP server view and confirm gavana is running with 11, 29, or 57 tools. Ask the agent to call guide_search — it requires no scope at all, so success confirms the transport independently of your permission choices.