documents/dev/code-server.md

code-servervscode
Table of Contents

code-server

Server config

~/.config/code-server/config.yaml — server host, port, and auth:

bind-addr: 0.0.0.0:8081
auth: password
password: PASSWORD
cert: false

User config

Directory: ~/.local/share/code-server/User/

settings.json

{
  "workbench.colorTheme": "Experimental Dark",
  "oaicopilot.baseUrl": "https://api.deepseek.com/v1",
  "oaicopilot.models": [
    {
      "id": "deepseek-v4-flash",
      "name": "DeepSeek V4 Flash",
      "provider": "DeepSeek",
      "owned_by": "deepseek",
      "include_reasoning_in_request": true,
      "thinking": { "type": "enabled" }
    },
    {
      "id": "deepseek-v4-pro",
      "name": "DeepSeek V4 Pro",
      "provider": "DeepSeek",
      "owned_by": "deepseek",
      "include_reasoning_in_request": true,
      "thinking": { "type": "enabled" }
    }
  ],
  "oaicopilot.retry": {
    "enabled": true,
    "max_attempts": 3,
    "interval_ms": 1000,
    "status_codes": []
  },
  "chat.viewSessions.orientation": "stacked",
  "workbench.startupEditor": "none",
  "chat.mcp.discovery.enabled": {
    "cursor-global": true
  },
  "editor.fontFamily": " Menlo, Monaco, Courier New, monospace",
  "vim.insertModeKeyBindings": [
    {
      "before": ["j", "j"],
      "after": ["<Esc>"]
    }
  ],
  "git.openRepositoryInParentFolders": "never"
}

mcp.json

File: ~/.local/share/code-server/User/mcp.json

{
	"servers": {
		"qwestly": {
			"type": "stdio",
			"command": "node",
			"args": [
				"/Users/dominick/Work/llm/qwestly-mcp-dev/src/index.js"
			]
		}
	},
	"inputs": []
}

OAI Compatible Provider for Copilot

Extension marketplace ID: johnny-zhao.oaicopilot — OAI Compatible Provider for Copilot by johnny-zhao.

Allows using any OpenAI-compatible API as a Copilot chat model provider. The oaicopilot.* settings in settings.json configure it:

  • oaicopilot.baseUrl — the API endpoint (e.g. https://api.deepseek.com/v1)
  • oaicopilot.models — list of model IDs to expose in the chat model picker

This is what enables deepseek-v4-flash and deepseek-v4-pro in the Copilot chat dropdown.

DeepSeek reasoning_content fix

Problem: In multi-turn conversations with DeepSeek models, the assistant's reasoning_content was dropped on subsequent requests because include_reasoning_in_request defaults to false. When thinking: { type: "enabled" } is set, DeepSeek's API requires that prior assistant reasoning_content be passed back in the conversation history. Without it, the API returns: "The reasoning_content in the thinking mode must be passed back to the API." (400 Bad Request)

Config fix (settings.json): Added "include_reasoning_in_request": true to both model entries.

Plugin patches (extension directory: ~/.local/share/code-server/extensions/johnny-zhao.oai-compatible-copilot-0.3.6/out/):

  1. provider.js:138-140 — Auto-detect when reasoning content must be preserved. If include_reasoning_in_request is not explicitly set, it now defaults to true when thinking.type === "enabled" or enable_thinking === true:

    includeReasoningInRequest: um?.include_reasoning_in_request ??
    (um?.thinking?.type === "enabled" || um?.enable_thinking === true),

  2. openai/openaiApi.js:98-100 — Removed the "Next step." fallback placeholder. Reasoning content is now only set when actual thinking content exists in the message, avoiding fabricated placeholders:

    if (modelConfig.includeReasoningInRequest && joinedThinking) {
    assistantMessage.reasoning_content = joinedThinking;
}

These patches make the explicit include_reasoning_in_request config setting technically optional — the plugin now infers it from the thinking configuration.

Important: After editing settings.json or plugin files, reload the code-server window (Command Palette → "Developer: Reload Window" or Cmd+R).

Other files in User/

  • caches/
  • globalStorage/
  • History/
  • workspaceStorage/
  • snippets/
  • machineid
  • customBuiltinExtensionsCache.json
  • systemExtensionsCache.json