Skip to main content
Anthropic ships Claude on three interfaces, the Claude Code CLI (and VS Code extension), the Claude Desktop app (Cowork), and Claude Web (claude.ai and the mobile apps), and sells it as Enterprise seats, Team seats, individual Pro / Max subscriptions, or plain API access with no subscription. What TrueFoundry can govern depends on the interface: Claude Code and Desktop can be pointed at a gateway, Web can’t. What it costs you depends on the plan, because routing through the gateway changes who pays Anthropic.

What you can govern, by interface and billing

Comparing Claude with Cursor, Codex, Copilot, and the other editors? The cross-tool table is on Integration with IDEs. This page goes one level deeper on Claude. How billing works for each column is in Seat-based or usage-based billing.
InterfaceEnterprise
(with API key)
Enterprise
(seat-based only)
Team · Pro · MaxOnly API key
(no Claude platform subscription)
Claude Code (CLI, VS Code)✅
Cost tracking, MCPs, Governance
✅
Cost tracking, MCPs, Governance
✅
Cost tracking, MCPs, Governance
✅
Cost tracking, MCPs, Governance
Claude Desktop (Cowork)✅
Cost tracking, MCPs, Governance
🟡
MCPs, Governance
🟡
MCPs, Governance
✅
Cost tracking, MCPs, Governance
Claude Web (claude.ai, mobile)🟡
MCPs, Governance
🟡
MCPs, Governance
🟡
MCPs, Governance
N/A. Web requires a paid plan

✅ Full: Cost tracking, MCPs, and Governance.
🟡 Partial: MCPs and Governance only; no Cost tracking. Inference stays on Anthropic.

  • Cost tracking: model calls go through the AI Gateway, so you get per-user usage, budgets, and rate limits.
  • MCPs: tool calls route through the MCP Gateway with allowlists, auth, and audit.
  • Governance: guardrails and policy checks on prompts and responses before they reach the model.

Seat-based or usage-based billing

Anthropic sells Claude two ways. Seat-based: Enterprise and Team seats, and Pro / Max subscriptions, bought on claude.ai, each with a usage allowance for the Claude apps. Usage-based: API keys from the Claude Console (or Claude on Bedrock, Vertex, or Microsoft Foundry through those clouds), billed per token. They are separate products with separate bills; a seat never covers API calls and an API key never draws on a seat’s allowance, even when both sit under one Enterprise contract. Which one the gateway bills depends on a single thing: what credential reaches Anthropic. Usage-based (API key mode). The app sends only your TrueFoundry key. The gateway identifies the user with it, then calls Anthropic with the API key stored on its model account. Anthropic sees an API call on that key and nothing else, so the user’s seat or subscription is never consulted, never drawn down, and never shown in the Claude admin console. Any plan, or no plan, behaves identically.
ClaudeClaude Code or Desktop
TrueFoundry key
▶
TrueFoundryTrueFoundryAI Gatewayidentifies the user, applies RBAC and budgets
API key on the gateway
▶
ClaudeAnthropicbills the API key per token
Seat-based (passthrough mode). The user signs in to Claude inside the app, so the app sends their Claude login token as well as your TrueFoundry key. The model account on the gateway has no API key, which tells the gateway to forward the Claude token untouched. Anthropic recognises the seat or subscription and bills it, exactly as if there were no gateway.
ClaudeClaude Code, signed in
Claude login tokenTrueFoundry key
▶
TrueFoundryTrueFoundryAI Gatewayidentifies the user, applies RBAC and budgets
Claude login token
▶
ClaudeAnthropicbills the seat or subscription
If you want usage billed against the seats or subscriptions you already pay for, configure passthrough mode. It exists for Claude Code only. Claude Desktop in third-party mode has no Claude login, so it is always usage-based; Claude Web has no gateway setting, so it is always seat-based. Both modes give the gateway full visibility: per-user attribution, budgets, rate limits, guardrails, and logs. The TrueFoundry key is present in both and is never billed; it only identifies the user. Configuration for each is under Authentication and who pays.

Get an API key and store it on the gateway

Usage-based billing needs an API key on the TrueFoundry model account. For Claude via Anthropic directly, that is an Anthropic API key. Keys are created in the Claude Console, which is separate from claude.ai: a Claude plan doesn’t come with one, and the admin console for Enterprise or Team only issues admin keys, which can’t call models.
1

Sign in to the Claude Console

Go to platform.claude.com and sign in, or create an organization. If your company already has one, ask its admin to add you rather than creating a second organization.
2

Create the key

Open Settings → API keys and click Create key. Name it for the gateway, set an expiration, and set Linked account to a service account rather than yourself, so the key keeps working if you leave the organization. If the button is disabled, your role doesn’t allow it; an organization admin can create a service account key for you. The key starts with sk-ant- and is shown only once, so copy it now. Anthropic’s walkthrough: Get your Claude API key.
3

Add billing

The Console bills per token against prepaid credits or an invoice, under Settings → Billing. A key with no credits returns errors on the first request.
4

Store it on the gateway

In TrueFoundry, go to AI Gateway → Models, add an Anthropic account, paste the key, and add the Claude models you want to expose. The key lives only on the gateway; users never see it, and every request is billed to it. Full steps in Anthropic model integration. For passthrough, create the account with the key left empty instead.
Claude models on Bedrock, Vertex, or Microsoft Foundry use that cloud’s credentials instead of an Anthropic key; see the Bedrock, Vertex, and Microsoft Foundry integrations.

Pick your starting point

Quick start

Route Claude Code on one machine through the gateway in five minutes.

Claude Code

CLI and VS Code extension, with per-plan authentication.

Claude Desktop & Cowork

Third-party inference mode, per user or fleet-wide.

Claude Web

Govern claude.ai and the mobile apps with inference hooks or aitori.

Govern MCP traffic

Route every MCP server Claude reaches through the MCP Gateway.

Enforce with MDM

Lock the gateway config on every developer machine.
Billing by plan, which token goes where, and the errors people hit most are answered in the FAQ.

Guardrails beyond the model call

Routing model traffic through the gateway governs what Claude sends to the model. Two hook integrations extend that to what Claude does:
  • Anthropic Inference Hooks: Enterprise only. One setting in Anthropic’s admin console sends every prompt from claude.ai, Desktop, Code, and the mobile apps to the gateway for an allow/deny verdict before inference. Nothing to install on devices.
  • Claude Code hooks: any plan. Client-side hooks post prompts and tool calls to the gateway, which can block or rewrite them.

Before you start

1

Create a TrueFoundry account

Sign up and follow the Gateway quick start to get your gateway base URL and API key.
2

Add a Claude model account

Connect Anthropic, Bedrock, or Vertex via Anthropic model integration. For subscription passthrough on Claude Code, leave the API key empty.
3

Pick the integration guide for your interface

Use the cards above, or start with the quick start.