For Claude Code
Roll it out to the fleet, not to one laptop.
Claude Code reads a managed-settings file before it reads anything a person typed. Push that file by MDM and two hundred laptops point at your endpoint without anyone opening a terminal.
{
"env": {
"ANTHROPIC_BASE_URL": "https://ai.acme.com",
"ANTHROPIC_MODEL": "claude-sonnet-4-5"
}
}Deploy the stack
One deployment into your own AWS account, in the region your Bedrock spend already sits in. Nothing is hosted by us.
Push the endpoint by MDM
The managed-settings file sets the base URL for the whole fleet. Engineers do not edit a shell profile, and cannot quietly opt out of it.
Each person gets a key
They sign in with Entra ID or Google Workspace and issue their own key, scoped by the budget and model policy an administrator already set.
Where the file goes
Same JSON on every platform. If your MDM can place a file, it can roll out TokenLock.
Or two lines, per person
For a pilot on one team, skip the MDM step entirely. Two environment variables and Claude Code is behind TokenLock.
No SDK, no proxy config, no change to how anyone works. The same two lines work for Claude Desktop and Cowork.
export ANTHROPIC_BASE_URL=https://ai.acme.com export ANTHROPIC_AUTH_TOKEN=tl_sk_...
What changes for the engineer
| Their experience | After TokenLock |
|---|---|
| Commands, agents, MCP servers | Unchanged. TokenLock speaks the same API. |
| Latency | One hop inside your own VPC, streaming preserved. |
| Reaching a budget | The reply carries a notice saying what happened and what to do. |
| Model choice | Whatever you allow, with a fallback you nominate. |
Point one team at it this week.
The first breakdown appears as soon as the first request lands. There is nothing to backfill and nothing to instrument.