LSP

Language Server Protocol (LSP) has been there way before AI arrived. This is exactly what’s being executed to get IntelliSense when developing within IntelliJ, Zed, VSCode, or other code editor.

When an agent edits code, it only sees a file and predicts how it should be written based on its understanding of the format (e.g. a .java, etc.), the project files (because the agent has read them), and the user’s request.

To counter potential invalid code and ease code navigation, LSPs can be connected to agent runtimes, and like MCP, they offer access to tools such as verifying that code compiles or finding what calls a function.

Combined with codebase indexing, an agent can navigate code far faster and find what it needs to implement its task(s).

repository/
└── .lsp.json
~/.claude/plugins/cache/<marketplace>/<plugin>/
└── .lsp.json

{
    "go": {
        "command": "gopls",
        "extensionToLanguage": { ".go": "go" },
        "startupTimeout": 120000
    }
}

repository/
└── .github/
    └── lsp.json
~/.copilot/
└── lsp-config.json
~/.copilot/installed-plugins/<marketplace>/<plugin>/
└── lsp.json

{
    "lspServers": {
      "go": {
          "command": "gopls",
          "fileExtensions": { ".go": "go" },
          "requestTimeoutMs": 90000
      }
    }
}