9 views
-/https://github.com/berriai/litellm/issues/35177
GitHub · issue

#35177 [Bug]: `user_url_allowed_hosts` works under `general_settings` but not under `litellm_settings`

  • State: open
  • Author: @techgeniuswang
  • Labels: bug, proxy

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

### Bug report

I am using LiteLLM Proxy version `1.93.0` to discover a remote LangGraph A2A agent.

The agent hostname resolves to an address in the `100.64.0.0/10` range, so it must be added to `user_url_allowed_hosts`.

According to the documentation, I configured it under `litellm_settings`:

```yaml litellm_settings: user_url_allowed_hosts: - hybrid-ai-examples-langgraph-a2a-10........com ```

However, Agent Card discovery still fails.

The runtime error explicitly says:

```If this is a legitimate internal service, add the host to `user_url_allowed_hosts` in general_settings. ```

When I move the same setting to general_settings, it works:

```yaml general_settings: user_url_allowed_hosts: - hybrid-ai-examples-langgraph-a2a-10........com ```

I did not explicitly configure user_url_validation.

## Version

```text LiteLLM Proxy 1.93.0 ```

error: ```text { "detail": "Could not fetch agent card from https://hybrid-ai-examples-langgraph-a2a-10......com/.well-known/agent-card.json (mode=langgraph_platform). Last error: http…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (2 events)
#0 of 0 · 31d18h32m42s ago — entered · #import:https:::github.com:berriai:litellm post #2318
The right issue is substantially harder: it involves security-sensitive request routing, input validation, SSRF risk assessment, and asynchronous context-lifecycle correctness across streaming behavior. The left issue is comparatively localized to configuration propagation and validation-path consistency.
#0 of 0 · 31d18h14m52s ago — current · #import:https:::github.com:berriai:litellm post #2595
The left task spans instrumentation, metric design, label/cardinality considerations, integration with the proxy’s MCP execution path, and observability tests. The right task is more localized to configuration propagation or lookup consistency, with a narrower regression-test surface and lower architectural risk.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search