LLM Agent Protocols/Memory

Integrating AI agents with databases today requires custom implementations for each framework—no standard exists for capability discovery, safe query execution, or persistent context across sessions. This project adds agent compatibility to Apache AsterixDB by implementing two emerging standards: the Model Context Protocol (MCP) for structured tool exposure and the Agent-to-Agent (A2A) protocol for multi-agent coordination, along with persistent agentic memory utilizing AsterixDB's native storage. Key deliverables include: McpServlet implementing JSON-RPC 2.0 protocol with tools/list, tools/call, and resources/read methods; ToolRegistry enabling dynamic capability discovery; QueryTool and SchemaTool exposing SQL++ execution and dataset introspection; rate limiting and query safeguards for production safety; AgentMemoryManager providing durable CRUD storage for session context via SQL++ datasets; MemoryServlet REST endpoints for memory operations; and A2A protocol endpoints for multi-agent task delegation (stretch goal). The implementation integrates with existing AsterixDB infrastructure—BasicAuthServlet for authentication, MetadataManager for schema access, and LSM storage for persistence—requiring no breaking changes. By project end, AI agents can discover AsterixDB's capabilities, execute queries like "find customers with orders over $1000," and recall previous interactions across sessions—transforming AsterixDB into an AI-native database that speaks the language of modern agent frameworks.