I'm getting API Error: 400 All target providers failed

[perdrix52]

I'm assuming I have done something wrong. I started Claude Code using:

ccr "Claude Code" with the following config.json

My config.json:
<
{
"LOG": true,
"LOG_LEVEL": "info",
"HOST": "127.0.0.1",
"PORT": 3456,
"API_TIMEOUT_MS": 600000,
"Providers": [
{
"name": "gemini",
"api_base_url": "https://generativelanguage.googleapis.com/v1beta/models/",
"api_key": "$GOOGLE_API_KEY",
"models": ["gemini-2.5-flash", "gemini-2.0-flash"],
"transformer": { "use": ["gemini"] }
}
],
"Router": {
"default": "gemini,gemini-2.5-flash",
"background": "gemini,gemini-2.5-flash",
"think": "gemini,gemini-2.5-flash",
"longContext": "gemini,gemini-2.5-flash",
"longContextThreshold": 60000
}
}

I got that error when I entered "status" to Claude Code.

What am I doing wrong please?

[slophisticated]

That message alone doesn't establish which part of the configuration is wrong. In current CCR, All target providers failed is an aggregate error; individual failures are stored in error.attempts (source). The outer 400 can therefore hide a different upstream status or a model-resolution failure.

Could you share your installed CCR version (npm ls -g @musistudio/claude-code-router --depth=0) and the failed request's resolved provider/model plus the error status/message? If available, include the relevant error.attempts entries. Please redact keys, Authorization headers, tokens, and private request content.

For the current 3.x UI, the documented path is Settings → Logs & Observability → Request logs, then open the failed request on Logs and inspect its response/error details (guide). Check the resolved model there rather than assuming the JSON shown is the active runtime configuration. Current releases persist application configuration in SQLite (CLI reference); the installed version matters when interpreting an older JSON configuration.

There is also a model-access detail worth checking: Google's current deprecation page lists gemini-2.0-flash as shut down, while gemini-2.5-flash has no announced shutdown but is limited to users who previously used the 2.5 models. Your shown routes target 2.5, so removing the unused 2.0 entry alone would not demonstrate a fix. The actual upstream error will distinguish model access, credentials, quota, or routing problems.

AI-assisted, source-based triage. I have not reproduced your live Gemini request.

[perdrix52]

`-- @musistudio/claude-code-router@3.1.3

I failed to locate the logs :( But I tried a "Check Connection" and got this message for both 2.0 and 2.5

HTTP 401: API keys are not supported by this API. Expected OAuth2 access token or other authentication credentials that assert a principal. See https://cloud.google.com/docs/authentication
gemini-2.0-flash

So I guess that I need to chose different models? Can you suggest what I should configure - I just lifted my config from a web page that suggested those settings

[slophisticated]

Thanks — the 401 narrows this down. It shows that authentication was rejected; it does not establish that choosing a different model will fix it.

For a starting setup in CCR 3.x, use the Google Gemini provider preset in Providers, rather than copying the older JSON example. CCR's documented Gemini preset uses:

• API base URL: https://generativelanguage.googleapis.com
• Protocol: Gemini Generate Content (gemini_generate_content)
• Provider credential: your Gemini API key issued through Google AI Studio, entered locally in the provider's API key field.

The preset base URL does not include /v1beta/models/. Also make sure this is the upstream Google credential, rather than a CCR client API key. Google's authentication guide documents both standard and authorization API keys; an OAuth-related error alone does not mean that the Gemini Developer API never accepts API keys.

Once the credential/endpoint combination is correct, select a currently available model from provider model discovery and test just that model. I would not recommend another fixed model ID before knowing what your account can access. CCR's provider guide describes the model selection and connection checks.

If that still returns 401, could you share only the configured base URL and protocol, and say whether the credential came from AI Studio, Vertex AI, or an OAuth login? Please keep the key/token itself private.

AI-assisted documentation/source review; I have not tested a live Google request or verified your account's model access.

[perdrix52]

Closing this - seems way too hard to set up - I think I'll just pay the Claude subscription.