Shutting Down MCP Server

[nimit2801]

{
  "mcpServers": {
    "google-calendar": {
      "command": "node",
      "args": [
        "/Users/nimitsavant/Desktop/nimit-devrev/google-calendar-mcp/build/index.js"
      ]
    }
  }
}

[mimi3421]

I think there is a Bug somewhere in the request POST JSON message accounting to the null string.

I tested the mcphost locally with the local llama.cpp server with empty mcpServer setting and sent a simple prumpt "hi". The log in mcphost was:

mcphost --model openai:phi4 --openai-url http://localhost:8080/v1 --openai-api-key 000000--debug
2025/04/15 00:29:32 INFO <cmd/root.go:487> Model loaded provider=openai model=phi4

You: hi
2025/04/15 00:29:34 DEBU <openai/provider.go:46> creating message prompt=hi num_messages=1 num_tools=0
2025/04/15 00:29:34 DEBU <openai/provider.go:55> converting message role=user content=hi is_tool_response=false
2025/04/15 00:29:34 DEBU <openai/provider.go:143> sending messages to OpenAI messages="[{Role:user Content:0xc000247560 ReasoningContent: FunctionCall: ToolCalls:[] Name: ToolCallID:}]" num_tools=0
2025/04/15 00:29:34 INFO <cmd/root.go:502> Shutting down MCP servers...
Error: error response with status 500

While the log in llama.cpp was:

srv log_server_r: request: POST /v1/chat/completions 127.0.0.1 500
srv log_server_r: request: {"model":"phi4","messages":[{"role":"user","content":"hi","reasoning_content":null},{"role":"user","content":"hi","reasoning_content":null}],"max_tokens":4096,"temperature":0.7}
srv log_server_r: response: {"error":{"code":500,"message":"Failed to parse messages: [json.exception.type_error.302] type must be string, but is null; messages = [\n {\n "role": "user",\n "content": "hi",\n "reasoning_content": null\n },\n {\n "role": "user",\n "content": "hi",\n "reasoning_content": null\n }\n]","type":"server_error"}}

I think the null in the request JSON should be "" which triggered the error in the JSON parser.

Edited:
I'm new to this and I think maybe the empty reasoning_content field should be omitted in the request POST and not just be set to null.

//pkg/llm/openai/types.go
type MessageParam struct {
	Role             string        `json:"role"`
	Content          *string       `json:"content"`
	ReasoningContent *string       `json:"reasoning_content",omitempty` // Add omitempty here
	FunctionCall     *FunctionCall `json:"function_call,omitempty"`
	ToolCalls        []ToolCall    `json:"tool_calls,omitempty"`
	Name             string        `json:"name,omitempty"`
	ToolCallID       string        `json:"tool_call_id,omitempty"`
}

Similar issues may be shown in link or link

[Waasi]

Pitching in just in case this is still an issue. For me, it turned out that the model I was using was not found. I got to this by logging the error returned here. I hope this helps.

[semidark]

Pitching in just in case this is still an issue. For me, it turned out that the model I was using was not found. I discovered this by logging the error returned here. I hope this helps.

I ran into the same issue, and at least in my case, @Waasi was correct. For me, not only was the model not found, but I also could not connect to the Ollama server. I had to correctly set my OLLAMA_HOST environment variable. After that, I got my first MCP server up and running.

So, I believe better logging when this error occurs would be really helpful.