Skip to main content
Connect Codex to Gcore MCP Server so Codex can call Gcore operations.

​
Step 1. Install Codex

Confirm the command is available:

​
Step 2. Sign in

Sign in through the browser.
Verify that Codex is signed in:

​
Step 3. Add Gcore MCP Server

The Codex configuration stores the variable name, not the token. Set GCORE_API_KEY in the shell that starts Codex. Enter the token at the prompt so the shell history does not store it. macOS uses Zsh, and Linux uses Bash. On both, the prompt hides the token. When a secret manager already exports GCORE_API_KEY, skip the prompt.
Open the Codex configuration and add this table. When the file already lists other servers, add the table without removing those entries.
Leave GCORE_TOOLS unset so the server stays in dynamic mode. Codex then receives six tools that reach every operation. Setting GCORE_TOOLS without GCORE_MCP_MODE selects direct mode. A value that matches nothing exits with status 2. Pattern rules are in Tool filtering.

​
Step 4. Verify the configuration

List the servers saved in the Codex configuration. This command does not start Gcore MCP Server or run an operation.
gcore appears as enabled. The Auth column says Unsupported because this server does not have its own sign-in. It reads GCORE_API_KEY from the environment. The command masks the key.
When gcore is missing, confirm that uv --version runs and that GCORE_API_KEY is set in the same shell. A Codex process started earlier does not see a variable exported later.

​
Step 5. Test with a query

Start Codex from that shell and ask:
Codex should use Gcore MCP Server to retrieve the DNS zones. When the connection works and the token has the required permissions, Codex returns the available zones or reports that none were found.
Requests that return “401 Unauthorized” indicate an invalid or expired token. Required roles are in API token permissions.