VS Code

Conifer models in the picker#

VS Code’s bring-your-own-key setting puts Conifer models in the same Chat model picker you already use. No Copilot plan is needed.

1. Open the Command Palette and run Chat: Manage Language Models.
2. Choose Add Models → Custom Endpoint and name the group “Conifer”.
3. When VS Code asks for the API type, pick Chat Completions. It then opens chatLanguageModels.json. Describe each model you want in the picker.

[
  {
    "name": "Conifer",
    "vendor": "customendpoint",
    "apiKey": "${input:coniferApiKey}",
    "apiType": "chat-completions",
    "models": [
      {
        "id": "claude-haiku-4-5",
        "name": "Claude Haiku (Conifer)",
        "url": "https://api.conifer.build/v1/chat/completions",
        "toolCalling": true,
        "maxInputTokens": 200000,
        "maxOutputTokens": 8192
      }
    ]
  }
]

The apiKey line is an input reference. VS Code prompts for the key and stores it itself, so the file carries no credential. Save the file and pick the model from the Chat picker like any other. Copy maxInputTokens and the capabilities from the catalog. VS Code uses them as the model’s context window.

The MCP server, for agent mode#

Create .vscode/mcp.json in the workspace. VS Code shows a Start action on the server entry and the tools appear in agent mode.

{
  "inputs": [
    {
      "type": "promptString",
      "id": "conifer-key",
      "description": "Conifer API key",
      "password": true
    }
  ],
  "servers": {
    "conifer": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "conifer-sdk", "conifer-mcp"],
      "env": { "CONIFER_API_KEY": "${input:conifer-key}" }
    }
  }
}

Keep the inputs block. VS Code prompts for the key once and stores it itself, so the file you commit carries no credential. The server runs from the published conifer-sdk package through npx, so there is nothing to build. The MCP server section lists each tool.

Which one to use#

Use BYOK when you want the session itself to run on a Conifer model. Use MCP when you want the agent, whatever model drives it, to consult the catalog, ask a cheaper model, compare candidates on a live prompt, or check the balance. Most setups end up with both.