constraint-mcp v2 Released

Existing AST (Abstract Syntax Tree)-based analysis had the limitation of only being able to catch structural rules such as import statements. The newly updated constraint-mcp v2 adds a Semantic Constraints layer using embedding models to inspect the actual intent of the code.

The key features are as follows:

• Domain Coherence: Checks whether a file's content is semantically consistent with the purpose of the module (e.g., authentication, JWT, etc.).
• Semantic Coupling Bans: Prevents a specific file from becoming semantically too close to a disallowed domain (e.g., including SQL queries in the API layer).
• Semantic Drift Detection: Detects when the nature of the code gradually changes compared to the initial baseline as code is written.

Technically, it uses fastembed and the BAAI/bge-small-en-v1.5 model to run locally at a fast speed of around 20ms even on CPU environments, adopting a fully offline approach that requires no API keys or cloud connection.