Connect Codex
Connect Krutrim Cloud MCP to Codex with the shared local Codex configuration. The ChatGPT desktop app, Codex CLI, and Codex IDE extension can use the same MCP server definition.
Before you begin
Follow the Overview to install the released package, sign in, and complete MFA if required.
Make sure
uvxis available on the local machine.Keep the IAM access token and refresh token in local protected configuration only.
1. Open the Codex configuration file
Open ~/.codex/config.toml. Create the .codex directory and configuration file if they do not already exist.
Scope: The configuration below runs Krutrim Cloud MCP locally through
stdio. Do not add actual tokens to a project configuration or commit them to source control.
2. Add the Krutrim Cloud MCP server
Add this entry to ~/.codex/config.toml. If you already have MCP servers, preserve the existing entries and add this one alongside them.
[mcp_servers.krutrim-cloud]
enabled = true
command = "uvx"
args = ["krutrim-mcp-server"]
[mcp_servers.krutrim-cloud.env]
KRUTRIM_ACCESS_TOKEN = "YOUR_IAM_ACCESS_TOKEN"
KRUTRIM_REFRESH_TOKEN = "YOUR_IAM_REFRESH_TOKEN"3. Check that Codex sees the server
In a terminal, run:
In the Codex CLI interactive interface, use /mcp to inspect server status. Start a new Codex session after changing the file. In the ChatGPT desktop app or IDE extension, save the configuration and restart the relevant surface.
4. Verify the package and Cloud access
Verify the local server before you invoke Cloud tools:
Then make a read-only request:
Review the tool call before allowing it. If tools are missing, verify the TOML section and environment variable names, then start a fresh Codex session. If the Cloud request is denied, complete MFA if required and validate IAM permissions and region.
Work safely
Use the safe operating workflow for destructive or mutating actions. Begin with a list operation, reuse the exact returned KRN or identifier, and request confirmation only after reviewing the proposed change.
For current Codex MCP product behavior and UI paths, see the OpenAI MCP documentation.
Last updated
Was this helpful?

