[BUG] OpenAI completion error: 500 Internal server error: Unsupported tool_call_parser

[MarkLiJing]

Problem (one or two sentences)

Date/time: 2025-12-20T06:57:07.662Z
Provider: openai
Model: DeepSeek-R1

500
OpenAI completion error: 500 Internal server error: Unsupported tool_call_parser: None (request id: 202512200652133627752312601517)

Context (who is affected and when)

It can not work

Reproduction steps

chat with Roo code

Expected result

it work well

Actual result

500 error ,it not works

Variations tried (optional)

No response

App Version

3.36.16

API Provider (optional)

None

Model Used (optional)

No response

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

I'd like to help with this issue, but I need some clarification to ensure I understand the root cause and can propose the right solution.

About the error:
The error message "Unsupported tool_call_parser: None (request id: ...)" is coming from the server you're connecting to, not from Roo-Code itself. This error format is typically from a LiteLLM proxy or similar middleware that doesn't know how to handle native tool/function calls for the DeepSeek-R1 model.

Could you please provide more details:

1. What backend/endpoint are you using? Are you connecting to:

   • DeepSeek's official API directly (api.deepseek.com)?
   • A LiteLLM proxy?
   • Azure AI Inference Service?
   • Another OpenAI-compatible proxy or service?

2. What is your base URL configuration? (You can redact any sensitive parts like API keys)

3. Have you tried using the native DeepSeek provider instead of the OpenAI-compatible provider? In Roo-Code settings, there's a dedicated DeepSeek provider option that handles DeepSeek models properly.

4. What tool calling mode are you using? Is native tool calling enabled for your mode, or are you using XML-based tool calling?

This information will help me determine whether this is:

• A configuration issue that can be resolved with settings changes
• A compatibility issue that requires code changes in Roo-Code
• An issue with the proxy/backend that needs to be addressed there

Please mention @roomote in your response with these details, and I'll be happy to propose a fix or provide guidance.

[MarkLiJing]

I'd like to help with this issue, but I need some clarification to ensure I understand the root cause and can propose the right solution.

About the error: The error message "Unsupported tool_call_parser: None (request id: ...)" is coming from the server you're connecting to, not from Roo-Code itself. This error format is typically from a LiteLLM proxy or similar middleware that doesn't know how to handle native tool/function calls for the DeepSeek-R1 model.

Could you please provide more details:

1. What backend/endpoint are you using? Are you connecting to:

   • DeepSeek's official API directly (api.deepseek.com)?
   • A LiteLLM proxy?
   • Azure AI Inference Service?
   • Another OpenAI-compatible proxy or service?

2. What is your base URL configuration? (You can redact any sensitive parts like API keys)

3. Have you tried using the native DeepSeek provider instead of the OpenAI-compatible provider? In Roo-Code settings, there's a dedicated DeepSeek provider option that handles DeepSeek models properly.

4. What tool calling mode are you using? Is native tool calling enabled for your mode, or are you using XML-based tool calling?

This information will help me determine whether this is:

• A configuration issue that can be resolved with settings changes
• A compatibility issue that requires code changes in Roo-Code
• An issue with the proxy/backend that needs to be addressed there

Please mention @roomote in your response with these details, and I'll be happy to propose a fix or provide guidance.

Same environment and configuration, version 3.36.14 has no problems.

Thanks for confirming that version 3.36.14 worked! This is very helpful.

I have identified the root cause: In version 3.36.15, a change was introduced (PR #10213) that merges native tool defaults (supportsNativeTools: true, defaultToolProtocol: "native") for the OpenAI Compatible provider. This means that if your custom model info for DeepSeek-R1 does not explicitly set defaultToolProtocol: "xml", the system now defaults to native tool calling, which causes the "Unsupported tool_call_parser" error with LiteLLM proxies.

I will prepare a fix that changes the default tool protocol for OpenAI Compatible providers from "native" to "xml", which will restore compatibility with third-party proxies like yours.

[MarkLiJing]

Thanks for confirming that version 3.36.14 worked! This is very helpful.

I have identified the root cause: In version 3.36.15, a change was introduced (PR #10213) that merges native tool defaults (supportsNativeTools: true, defaultToolProtocol: "native") for the OpenAI Compatible provider. This means that if your custom model info for DeepSeek-R1 does not explicitly set defaultToolProtocol: "xml", the system now defaults to native tool calling, which causes the "Unsupported tool_call_parser" error with LiteLLM proxies.

I will prepare a fix that changes the default tool protocol for OpenAI Compatible providers from "native" to "xml", which will restore compatibility with third-party proxies like yours.

Thanks

I have opened PR #10260 with an attempt to fix this issue.

The fix changes the default tool protocol for the OpenAI Compatible provider from "native" to "xml", which should restore compatibility with third-party proxies like LiteLLM.

Feedback welcome!

[MarkLiJing]

I have opened PR #10260 with an attempt to fix this issue.

The fix changes the default tool protocol for the OpenAI Compatible provider from "native" to "xml", which should restore compatibility with third-party proxies like LiteLLM.

Feedback welcome!

I’m very sorry—here’s an update. We verified version 3.36.14 and it still has problems. Version 3.36.10 works fine. The API provider is OpenAI Compatible.

[daniel-lxs]

Hey, thank you for reporting, I'll take a look