Claude for Excel & PowerPoint

Use Portkey as the LLM gateway for Claude for Excel and Claude for PowerPoint add-ins — with observability, governance, and multi-provider routing.

Claude for Excel and Claude for PowerPoint are Microsoft Office add-ins that let users chat with their spreadsheets and presentations using Claude. Both add-ins support an Enterprise Gateway mode — instead of signing in with a Claude account, users point the add-in at your organization’s LLM gateway.

Portkey works as a drop-in gateway for this setup. It exposes the Anthropic Messages API format the add-ins require, and adds observability, budget controls, and multi-provider routing on top.

Cost Visibility

See exactly what each user, team, or department is spending through the add-ins. Set hard budget limits before bills surprise you.

Governance & Access Control

No raw API keys handed to employees. One Portkey API key controls access. Revoke or rotate centrally.

Multi-Provider Routing

Route to Anthropic directly, AWS Bedrock, or Google Vertex AI — switch providers with a single config change.

Full Request Logs

Every prompt and response logged in Portkey. Search, filter, and audit usage across your organization.

How It Works

The add-ins connect to any Anthropic Messages API-compatible endpoint. You give each user:

• Gateway URL: https://api.portkey.ai
• API token: A Portkey API key (optionally scoped with a Config)

Claude for Excel/PowerPoint → Portkey Gateway → Anthropic / Bedrock / Vertex AI

Portkey exposes the two endpoints the add-ins require:

Endpoint	Purpose
POST /v1/messages	Send messages to Claude (streaming + non-streaming)
GET /v1/models	Model discovery for the add-in login flow

Setup

Step 1: Add Your Provider in Model Catalog

1. Go to Model Catalog → Add Provider
2. Select your provider — Anthropic, AWS Bedrock, or Google Vertex AI
3. Enter your credentials and save with a slug (e.g., anthropic-prod)

Model Catalog Setup Guide →

Full instructions for adding providers and managing credentials

Step 2: Create a Config (Recommended)

Create a Config to set the default provider and model, and to forward the required headers. This ensures the add-in works correctly without any additional configuration from end users.

{
  "provider": "@anthropic-prod",
  "forward_headers": ["anthropic-beta", "anthropic-version"]
}

Forwarding anthropic-beta and anthropic-version headers is required. The add-ins use these headers to enable beta features and specify the API version. If your gateway doesn’t forward them, the add-ins may not work correctly.

Save the Config and note the Config ID.

Step 3: Create a Portkey API Key

1. Go to API Keys → Create new key
2. (Optional) Attach your Config from Step 2
3. (Optional) Set a budget limit for this key
4. Copy the API key

Step 4: Configure the Add-In

Share the following credentials with your users:

• Gateway URL: https://api.portkey.ai
• API token: The Portkey API key from Step 3

In Excel or PowerPoint:

1. Open the Claude add-in
2. On the sign-in screen, select “Enterprise gateway”
3. Enter the Gateway URL and API token
4. Click Test connection — the add-in validates the connection and discovers available models

Credentials are stored locally in the add-in’s browser localStorage. They are never sent to Anthropic. Only distribute gateway-issued tokens — not raw cloud provider credentials.

Provider Configurations

Route to Anthropic Directly

{
  "provider": "@anthropic-prod",
  "forward_headers": ["anthropic-beta", "anthropic-version"]
}

Set up @anthropic-prod in Model Catalog with your Anthropic API key.

Route to Amazon Bedrock

{
  "provider": "@bedrock-prod",
  "forward_headers": ["anthropic-beta", "anthropic-version"]
}

Set up @bedrock-prod in Model Catalog with your AWS credentials. Portkey handles the translation from Anthropic Messages API format to Bedrock’s InvokeModel format.

Route to Google Vertex AI

{
  "provider": "@vertex-prod",
  "forward_headers": ["anthropic-beta", "anthropic-version"]
}

Set up @vertex-prod in Model Catalog with your GCP service account credentials.

Multi-Provider Fallback

For high availability, configure automatic fallback across providers:

{
  "strategy": { "mode": "fallback" },
  "targets": [
    {
      "provider": "@anthropic-prod",
      "forward_headers": ["anthropic-beta", "anthropic-version"]
    },
    {
      "provider": "@bedrock-prod",
      "forward_headers": ["anthropic-beta", "anthropic-version"]
    },
    {
      "provider": "@vertex-prod",
      "forward_headers": ["anthropic-beta", "anthropic-version"]
    }
  ]
}

If Anthropic’s API is down, requests automatically route to Bedrock — users never see an error.

Enterprise Governance

Budget Controls Per Team

Create separate API keys for different teams, each with their own budget:

1. Go to API Keys → Create key for each team
2. Attach a budget limit (e.g., $500/month per team)
3. Distribute the team-specific key to that team’s employees

When a team hits their budget, their key stops working — other teams are unaffected.

Workspace Isolation

Use Portkey Workspaces to fully isolate teams. Each workspace gets its own:

• Logs and analytics
• Budget limits
• API keys and access controls

Add Metadata for Chargeback

Add metadata to track usage by department, cost center, or user. Attach this to your Config or API key:

{
  "provider": "@anthropic-prod",
  "metadata": {
    "department": "finance",
    "cost_center": "CC-1234"
  },
  "forward_headers": ["anthropic-beta", "anthropic-version"]
}

Filter logs in the Portkey Dashboard by any metadata field.

What’s Supported Through the Gateway

Feature	Available
Chat with spreadsheet / deck	✓
Read and edit cells, slides, formulas	✓
Streaming responses	✓
Web search (Vertex AI)	✓
Connectors (S&P, FactSet, etc.)	—
File uploads	—
Memory	—
Skills	—
Working across apps	—

Connectors, file uploads, memory, and skills require a Claude account sign-in. These are not available through any LLM gateway (including Portkey). If your users need these features, they should sign in with a Claude account instead.

Troubleshooting

Symptom	Cause	Fix
”Connection refused” or network error	Gateway URL unreachable	Verify users can reach https://api.portkey.ai — check firewall/VPN rules
401 Unauthorized	Invalid or expired API key	Regenerate the Portkey API key and redistribute
403 Forbidden	Provider credentials lack permissions	Check your provider credentials in Model Catalog
404 Not Found	Wrong base URL format	Use https://api.portkey.ai — do not append /v1/messages
”No models available”	/v1/models not returning results	Models are discovered automatically via Portkey — verify your provider slug is correct
Streaming fails or hangs	SSE not passing through	Portkey supports SSE natively — check if a corporate proxy is stripping SSE connections
Requests not appearing in logs	API key misconfigured	Ensure the API key attached to the Config is a valid Portkey key
Features missing (Connectors, Skills)	Gateway limitation	These require Claude account sign-in — see the feature comparison

Next Steps

Gateway Configs

Learn about fallbacks, load balancing, and routing strategies

Budget & Rate Limits

Set spending controls per team or user

Observability

Logs, traces, and analytics for all add-in requests

Workspaces

Isolate teams with separate budgets and access controls