# LiteLLM Claude Code Proxy
Open-source [[LiteLLM]] configuration that lets [[Claude Code]] run against an Azure AI Foundry deployment instead of Anthropic's hosted API. Published by National Bank of Belgium. It's a thin wrapper: a `litellm_config.yaml` with a wildcard model route, a `start_proxy.sh` launcher, a Dockerfile, and a `.env` template. Any model name Claude Code sends is forwarded to one Azure deployment (for instance `azure/responses/gpt-5.3-codex`).
## Why it exists
Claude Code speaks Anthropic's `/v1/messages` API. Enterprises often can't use Anthropic directly but have an Azure AI Foundry tenant with approved models. LiteLLM translates between Anthropic's format and Azure's `responses` / Chat Completions API, so Claude Code keeps working against a corporate-approved backend.
## How it works
- `model_name: "*"` wildcard in `litellm_config.yaml` — every model name is routed to a single Azure deployment
- Claude Code points at the proxy via two env vars:
- `ANTHROPIC_BASE_URL=http://localhost:4000`
- `ANTHROPIC_AUTH_TOKEN=<LITELLM_MASTER_KEY>`
- `LITELLM_LOCAL_MODEL_COST_MAP=true` skips the remote cost map fetch, avoiding SSL errors behind corporate proxies
- `start_proxy.sh` uses `uv run` with `UV_NATIVE_TLS=true` to use the system certificate store — another corporate-proxy workaround
## Three ways to configure Claude Code
- **Env vars per session** — `export ANTHROPIC_BASE_URL=...` before launching `claude`
- **Persistent** — add an `env` block to `~/.claude/settings.json`
- **VS Code** — add a `claude-code.env` block to VS Code `settings.json`
## Docker
Uses the official `docker.litellm.ai/berriai/litellm:main-stable` image. Two options: build the included Dockerfile, or mount `litellm_config.yaml` directly into the upstream image. Configuration stays in `.env`.
## Takeaways
- Clean pattern for using Claude Code with enterprise-hosted non-Anthropic models ([[Bring Your Own Key (BYOK)]] inverted: bring your own model, keep the client)
- Same pattern works for any Anthropic-compatible client — not just Claude Code
- Practical reference for corporate SSL-proxy workarounds in [[LiteLLM Proxy Configuration]]
- Complements [[Claude Max API Proxy]], which goes the other direction (Max subscription → OpenAI-compatible API)
## References
- https://github.com/NationalBankBelgium/litellm-claude-code-proxy
- https://docs.litellm.ai/docs/tutorials/claude_non_anthropic_models
- https://docs.litellm.ai/docs/proxy/docker_quick_start
## Related
- [[LiteLLM]]
- [[LiteLLM Proxy Configuration]]
- [[Claude Code]]
- [[Claude Code via GitHub Copilot]]
- [[Claude Max API Proxy]]
- [[AI Gateway]]
- [[Anthropic]]