Claude for Excel and Claude for PowerPoint can work through an organization’s internal LLM gateway connected to AWS Bedrock, Google Cloud Vertex Artificial Intelligence, or Microsoft Azure, allowing use without a Claude account. Organizations already using the same gateway pattern for Claude Code can point the Office add-ins at that existing endpoint, so no new infrastructure is required. End users need the add-in installed, plus a gateway URL and API token supplied by their IT team.
After an administrator deploys the add-ins, users open Excel or PowerPoint, launch the Claude add-in, choose “Enterprise gateway,” and enter the provided URL and token. The add-in uses the token in the Authorization: Bearer header with every request and verifies the connection before loading the main experience. Credentials are stored locally in browser localStorage inside the add-in’s sandboxed iframe and are not synced to Anthropic’s servers. Because the Office add-in runs in a sandboxed iframe, it cannot use the OS keychain the way Claude Code does, so organizations should provide gateway-issued tokens rather than raw cloud provider credentials.
For IT teams, the Office add-ins support only the Anthropic Messages API format via the Unified endpoint. Your gateway must expose these two endpoints: /v1/messages and /v1/models. The gateway must also forward the anthropic-beta and anthropic-version headers, which are required for add-in functionality. On login, the add-in validates the token via GET /v1/messages and then attempts to discover available Claude models via GET /v1/models. If the gateway does not provide a model list there, the add-in falls back to manual model ID entry.
The gateway-based setup differs from Claude Code in several areas. Claude Code supports additional API formats, more flexible credential handling, token rotation helpers, and configurable custom model names, while the Office add-ins rely on unified endpoints, manual credential entry, and browser-based storage. Anthropic also outlines example LiteLLM routing configurations for Anthropic, Bedrock, Vertex, and Azure, with model examples such as claude-sonnet-4-5-20250929.
Feature support is narrower through an enterprise gateway than through a direct Claude account sign-in. Chat with spreadsheets or decks and editing cells, slides, and formulas are available in both modes, while connectors are marked Coming soon, working across apps shows ✓ and -, skills are marked Coming soon, file uploads show ✓ and -, and web search is listed as Vertex only. Anthropic says it does not have access to a customer’s AWS, Google, or Microsoft instance, including prompts or outputs it contains, and does not train generative models on that content, though it can access metadata such as tool use and token counts for analytics and product improvement. Troubleshooting guidance covers connection failures, permission errors, incorrect base URLs, missing model lists, and SSE pass-through problems for streaming responses.
