lai-gen-mcp

Generate MCP Server

This guide instructs an agent on how to generate a Model Context Protocol (MCP) server for an arbitrary Langium-based DSL. The generated server exposes a validate tool that accepts DSL source code and returns diagnostics (errors, warnings, hints, information) by running the input through the DSL's actual Langium parser and validator. Any MCP-compatible client (Claude Code, Cursor, VS Code, etc.) can then validate DSL code on the fly.

You may also use the lai and langium skills to achieve a better understanding of langium-ai and the Langium-based DSL in question.

When to Use

• After a Langium project has been initialized with lai init and a descriptor exists
• When you want to give an MCP-compatible AI assistant the ability to validate code written in your DSL
• When building a feedback loop where an LLM generates DSL code and can self-check it via MCP

Prerequisites

The target Langium project must have:

1. A working Langium grammar (.langium file) and generated TypeScript artifacts (ast.ts, grammar.ts, module file)
2. A service creation function — every Langium project has a create<Name>Services function in its module file (e.g., createMyDslServices in src/my-dsl-module.ts). This is the entry point for all language services.
3. Node.js and npm available in the project

Inputs

Gather the following from the project before generating:

1. Language descriptor (language.descriptor.yml) or lai.config.jsonc — to find the grammar path, service paths, and language name
2. The module file — contains the create<Name>Services(...) function. Typically located at src/<name>-module.ts or src/language/<name>-module.ts. The descriptor's services.module field points to it if present; otherwise search for createServices or create.*Services in the src/ directory.
3. The language metadata — the LanguageMetaData export (usually in generated/ast.ts or generated/module.ts) tells you the language's file extensions