# Grodmir — Claude Code gateway and prompt data protection

> Route Claude Code model requests through a self-hosted Grodmir gateway. Review the endpoint and virtual-key setup, optional repository hooks, and the separate connection needed for MCP tool traffic.

Grodmir is an AI gateway for data protection and AI usage governance in organisations.

## How does Claude Code use the AI gateway?

Grodmir is an AI gateway for Claude Code when ANTHROPIC_BASE_URL points to the customer gateway and ANTHROPIC_API_KEY holds a Grodmir virtual key. Provider keys stay in the gateway.

[Answer source](https://grodmir.com/en/integrations/claude-code#how-does-claude-code-use-the-ai-gateway)

## What was tested and what needs checking?

The repository guide records Claude Code tested against a live gateway. Optional repository hooks refuse excluded paths; long system prompts can trigger local-model false positives. Signed thinking blocks are immutable; a required rewrite is blocked rather than forwarded.

[Answer source](https://grodmir.com/en/integrations/claude-code#what-was-tested-and-what-needs-checking)

## Does the Claude Code gateway also protect MCP servers?

Pointing Claude Code at the model gateway does not proxy its direct MCP server connections. Configure each intended MCP server through the Grodmir MCP proxy or central MCP gateway to inspect tool arguments and results on that connection. Optional Claude Code repository hooks serve a different purpose: they refuse supported file and shell operations that touch excluded paths.

[Answer source](https://grodmir.com/en/integrations/claude-code#mcp-coverage)

## How should a team verify Claude Code data protection?

Check that ANTHROPIC_BASE_URL names the customer gateway and ANTHROPIC_API_KEY contains a Grodmir virtual key rather than a provider credential. In a pilot, use synthetic examples to check allowed requests, policy blocks and any required pseudonymisation, then inspect audit metadata without recording prompt content. Enforce mode applies policy blocks and transformations; observe mode forwards content unchanged. Detection false positives and missed data still need evaluation on the intended workflow.

[Answer source](https://grodmir.com/en/integrations/claude-code#verify-protection)

- [Grodmir — self-hosted AI gateway](https://grodmir.com/en/ai-gateway)
- [Grodmir — Prompt DLP for sensitive AI data](https://grodmir.com/en/prompt-dlp)
- [Grodmir — MCP proxy for tool data protection](https://grodmir.com/en/mcp-proxy)
- [Grodmir — self-hosted MCP gateway with SSO and tool policies](https://grodmir.com/en/mcp-proxy/gateway)
- [What do the Grodmir categories mean?](https://grodmir.com/en/glossary)
- [Grodmir — AI gateway selection criteria](https://grodmir.com/en/how-to-choose-an-ai-gateway)

## How can you verify this behaviour?

- [claude-code (EN)](https://grodmir.com/dokumentacja/zrodla/claude-code.md)
- [repo-rules (EN)](https://grodmir.com/dokumentacja/zrodla/repo-rules.md)

Content updated: 2026-11-14

Protection covers connected-channel traffic only; detection can miss data and raise false alarms. [Grodmir — AI gateway protection limits](https://grodmir.com/en/limitations)

Source: https://grodmir.com/en/integrations/claude-code
