Description

When starting VS Code with MCP enabled, it always tries to run the following command:

docker run -i –rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server

and shows the error:

The command “docker run -i –rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server” needed to run GitHub was not found.

What I tried

• Docker is installed and works fine (docker --version → 28.3.0).
• Running the exact same docker run ... command manually in the terminal works correctly.
• I already set GITHUB_PERSONAL_ACCESS_TOKEN in my shell (~/.zshrc), and verified it is available in the terminal.
• I checked both settings.json and .vscode/mcp.json in my workspace, but I cannot find where this default command is being set.
• I even tried rewriting my mcp.json with "command": "docker" and "args": [...], but the error persists.

Problem

It seems VS Code (or the Copilot MCP integration) is hardcoding docker run ... as a single command string, instead of command: "docker" + args: [...].
Because of this, it cannot find the command, even though Docker is correctly installed.

Expected behavior

• VS Code should either:
  • Use the configuration from mcp.json without overriding it, OR
  • Correctly split docker into command and args instead of treating the whole docker run ... as a single executable.

Environment

• macOS (Apple Silicon)
• Docker Desktop 28.3.0
• VS Code latest (with Copilot + MCP enabled)
• github-mcp-server v0.12.1

Question

How can I stop VS Code from auto-injecting this broken "docker run ..." command, and instead use the correct configuration from my mcp.json?
Is this a bug in the current github-mcp-server integration?