Skip to main content
Make a personal key under AI Gateway, My keys, which also shows the settings below under Use it in your coding agent, with your organization’s address filled in. Use it for your own coding agent, so its traffic, cost and tool use are yours. Your organization’s guardrails, budgets and model access apply to every request it sends. A coding agent puts its own instructions into each request, so it is refused when Prompt Injections is set to Block in your organization’s guardrails: keep that guardrail on Log only there.

Claude Code

Add to the env block of ~/.claude/settings.json:
To use your organization’s MCP servers in Claude Code, add Altrum’s MCP endpoint. The endpoint answers only a request that carries a key, and shows only the tools that key may use. For Codex CLI, Cursor and VS Code, see MCP tools.

Codex CLI

Set your key, then add a provider to ~/.codex/config.toml:
Choose any model your key allows with --model, for example --model anthropic/claude-sonnet-4-5. Codex warns that it has no details for a model it does not know; the request still works.

Gemini CLI

When Gemini CLI asks how to sign in, choose the Gemini API key.

Cursor

In Cursor’s model settings, enter your personal key as the OpenAI API key, turn on Override OpenAI Base URL, and enter https://api.altrum.ai/cursor. Add the models you want as provider/model. To use your organization’s MCP servers, add Altrum’s MCP endpoint, https://api.altrum.ai/mcp, as an HTTP MCP server with your personal key in the Authorization header.

When the MCP endpoint refuses

A client that sends no key gets: