[BUG] Tools can't be called

[farazoman]

Problem (one or two sentences)

When the LLM wants to run a tool, the output is shown in xml without calling the tool.

Context (who is affected and when)

Happens always across different repos.

See example output I see in roocode extension:

<ask_followup_question>
<question>Hello! I'm here to help you with the tokens-api project. What would you like to know or work on?</question>
<suggestions>
<suggestion>Explain the architecture and how token generation works</suggestion>
<suggestion>Help me understand the Lambda handlers and their responsibilities</suggestion>
<suggestion>Show me how the CDK infrastructure is organized</suggestion>
<suggestion>Explain the data pipeline and Airflow integration</suggestion>
</suggestions>
</ask_followup_question>

I'll analyze the codebase to provide you with a comprehensive overview. Let me start by examining key files to understand the architecture.

<read_file>
<path>ReadMe.md</path>
</read_file>

Reproduction steps

1. open a chat with claude. Ask it anything, just hi for example
2. see the response

Expected result

Tools can be called and the tool execution xml should not be shown in raw form

Actual result

Tools aren't called, and xml is shown for tool calls

Variations tried (optional)

No response

App Version

3.36.13

API Provider (optional)

Anthropic

Model Used (optional)

claude sonnet 4.5, also others, was getting api error with google pro and opus

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[mrubens]

Hi Faraz, can you try upgrading to a more recent version of the app? We've fixed a lot of issues related to this.

[farazoman]

@mrubens I've just updated and now I get this error:

Date/time: 2025-12-22T16:43:12.593Z
Extension version: 3.36.16
Provider: litellm
Model: Gemini 3 Pro

LiteLLM streaming error: 400 litellm.UnsupportedParamsError: `parallel_tool_calls=False` is not supported by Gemini when multiple tools are provided. Specify a single tool, or set `parallel_tool_calls=True`. If you want to drop this param, set `litellm.drop_params = True` or pass in `(.., drop_params=True)` in the requst - https://docs.litellm.ai/docs/completion/drop_params. Received Model Group=Gemini 3 Pro
Available Model Group Fallbacks=None

We're using litellm actually, not claude directly. Is this likely a litellm issue?

Thanks in advance

[mrubens]

@farazoman ok we're going to try #10274 to see if that fixes things for you

[farazoman]

@mrubens thanks, also have some engineers on our side seeing if there's some config on our llm provider they can fix. Appreciate the fast triage!

[mrubens]

@farazoman would any of you be up for seeing if it's fixed in the Roo Code Nightly version of the extension?

[farazoman]

@mrubens I've confirmed on the nightly I don't see issues anymore, I will be using that for now. Our AI engineers are reviewing what is causing this issue on our side to put a fix in as well. Will let you know once they make their updates

[farazoman]

@mrubens our team has also fixed the issue on the api calls, apparently that flag isn't fully supported yet, but they made it so that using unsupported flags don't cause issues. Closing this, thanks again on the fast feedback