MCP server

The Gigadrive Network MCP server lets an AI assistant such as Claude deploy, configure and debug your applications on your behalf. Connect it when you want to ask an assistant why a deployment failed, which requests are erroring, or to change an environment variable without opening the console.

The assistant acts as you. Every tool call goes through the same permission checks as the API reference, so it can reach only the organizations and applications you can reach.

Connect Claude

On the Free plan you can add one custom connector; paid plans have no limit. On Team and Enterprise plans an owner adds the connector under Organization settings > Connectors, and each member then connects their own account.

1. In Claude, open Customize > Connectors.
2. Select Add, then Add custom connector.
3. In Connector name, enter Gigadrive Network.
4. In Remote MCP server URL, enter https://api.gigadrive.network/mcp, then save the connector. Leave the client ID and secret empty: Claude registers itself with Gigadrive.
5. Connect the connector. Claude opens the Gigadrive sign-in page.
6. Sign in and review the permissions, then select Allow.

Claude lists the Gigadrive Network tools once the connection succeeds. Ask it something like "Why did the last deployment of my app fail?" to check that it works.

Connect Claude Code

Add the server from your terminal:

claude mcp add --transport http gigadrive-network https://api.gigadrive.network/mcp

Start Claude Code, run /mcp, select gigadrive-network, and choose Authenticate. Your browser opens the Gigadrive sign-in page. On a machine without a browser, run claude mcp login gigadrive-network --no-browser instead.

Connect another client

Any MCP client that supports remote servers with OAuth can connect, including Cursor, VS Code and ChatGPT. Add https://api.gigadrive.network/mcp as a remote or HTTP server in the client's MCP settings. The client registers itself with Gigadrive and opens the sign-in page; no client ID or secret is needed.

Connect with an API key

For CI and other unattended use, authenticate with an API key instead of signing in. The assistant then acts as the key's application rather than as you, and the server offers only the tools the key's scopes allow.

Tokens minted from a key expire after five minutes, so the client must fetch a fresh one when it connects. In Claude Code, a headersHelper command does this: Claude Code runs it on every connection and again when the server rejects a token. Save this script as /usr/local/bin/gigadrive-mcp-headers.sh and make it executable with chmod +x. It reads the key from GIGADRIVE_CLIENT_ID and GIGADRIVE_CLIENT_SECRET and needs curl and jq:

#!/bin/sh
# Prints the Authorization header Claude Code sends to the Gigadrive Network MCP server.
token=$(curl -sS -X POST https://api.gigadrive.network/oauth2/token \
  -d grant_type=client_credentials \
  -d client_id="$GIGADRIVE_CLIENT_ID" \
  -d client_secret="$GIGADRIVE_CLIENT_SECRET" | jq -r .access_token)
printf '{"Authorization":"Bearer %s"}' "$token"

Then register the server with the helper in your project's .mcp.json:

{
  "mcpServers": {
    "gigadrive-network": {
      "type": "http",
      "url": "https://api.gigadrive.network/mcp",
      "headersHelper": "/usr/local/bin/gigadrive-mcp-headers.sh"
    }
  }
}

Clients without such a hook can send a static Authorization: Bearer header, which stops working after five minutes.

Permissions

Signing in grants read access to your profile, organizations, applications, deployments, and request logs, plus offline_access, so the assistant stays connected without signing you in again. The first time the assistant calls a tool that needs more, such as creating a deployment or changing an environment variable, your client asks you to approve that extra access. The approval keeps everything you granted earlier. If you decline, that tool stays unavailable and the rest keep working.

Tools for actions you are not allowed to perform still fail with a permission error, which the assistant reports back to you.

The connection lasts as long as the assistant uses it at least once every 30 days. To end it, remove the connector in your client.

AI Gateway budget, policy and usage tools are not available through sign-in. They appear only for an API key that carries the matching network:ai_gateway:* scopes. See Scopes.

Tools

Each tool tells the assistant whether it only reads, changes something, or destroys data. Clients use this to ask for your confirmation before a destructive call, and the server instructs the assistant to confirm those with you first.

The server also offers two prompts: debug_application investigates recent errors of an application, and deploy_application deploys a Git ref and reports the result.

Sensitive environment variable values are never returned. Secrets from create_api_key and create_s3_credential appear once in the conversation, so store them right away.

Related pages

• API reference
• Scopes
• Observability