How to use gopls MCP with VS Code

The Go team recently added an official Model Context Protocol (MCP) server to gopls, the language server protocol (LSP) implementation for Go. This allows IDEs and other AI coding agents to directly access some features of the LSP. While most of the features provided by the MCP already exist in AI coding tools, this can improve the efficiency of token use by avoiding searching and reading whole files.

Enable gopls MCP in VS Code

Now, let's get to the reason you're here! With a few simple steps, we can start using this new MCP feature in VS Code:

1. Install/update gopls

go install golang.org/x/tools/gopls@latest

• I currently have Go 1.24.0 with v0.20.0 of gopls, but newer versions exist at the time of writing this

2. Enable VS Code to run the server

VS Code automatically runs the gopls server for Go projects. A simple update to the config will also start the MCP server.

• Navigate to Settings and find Go: Language Server Flags
• Click Edit in settings.json and add:

  "go.languageServerFlags": [
      "-mcp.listen=localhost:8092",
  ]

• Restart/reload VS Code

3. Enable the MCP server for agents

• Use the MCP: Add Server... command
• Select HTTP
• Input the address: http://localhost:8092
• Alternatively, just edit .vscode/settings.json with this content:

  {
    "servers": {
      "gopls": {
        "url": "http://localhost:8092",
        "type": "http"
      }
    }
  }

4. Setup instructions for using the tools

The Go team provides a base instruction to improve how agents use the feature. Setting up VS Code to use this feature will greatly improve how it uses the MCP tools.

• First, create the instructions directory

  mkdir -p .github/instructions

• Then, start by telling VS Code to use this for Go files:

  cat <<EOF > .github/instructions/gopls.instructions.md
  ---
  applyTo: "**/*.go"
  ---
  EOF

• Finally, add the gopls instruction:

  gopls mcp -instructions >> .github/instructions/gopls.instructions.md

Use it!

Now you should have gopls MCP setup in your Go project! Open up the agent chat and start coding!

The simplest place to start is to ask about references for a function or type in your code: "Describe where MyFunction is used."

This should use the go_symbol_references tool to find references to this function.

Why MCP?

One of the most impactful features is the ability to use the go_diagnostics tool. Besides the usual information about diagnostic issues, this also allows the agent to automatically review and apply fixes suggested by the LSP. This greatly improves consistency and accuracy of code fixes since it uses deterministic analysis provided directly by the Go maintainers.

If you are regularly programming with newer versions of Go, you might be familiar with the "modernize" suggestions. I most-commonly see these with typical for i := 0; i < 10; i++ loops that could use the new for i := range 10 syntax. This feature is a big win for the language in the world of AI coding. By definition, LLMs always output "old" code since they are trained on existing codebases. This means any new language features won't be used by LLMs until enough human coders have introduced them into the training data.

The new analyze and modernize tools allow coding agents to automatically update to new language features that they don't even know about yet. This improves the overall quality of a codebase even when using AI agents. This is an important feature helping to keep Go relevant as software development is rapidly changing.

To demonstrate this tool, write a simple function using a for loop:

func Example() {
    for i := 0; i < 10; i++ {
        println("Hello, World!")
    }
}

Then, ask the agent to run diagnostics on it! It should look something like this: