Java Agent CLI (Claude Code-like) — AI agent execution, planning, MCP tool ecosystem, and secure browser automation
Built and implemented an LLM-driven command-line AI agent workflow that supports tool execution and debugging. The system uses MCP (Model Context Protocol) with JSON-RPC to enable structured function calling and multi-tool orchestration. It focuses on agent planning, execution modes, and secure browser automation integration for reliable AI-assisted development. • Implemented an MCP 2025-03-26 compatible JSON-RPC 2.0 client with async request/response pairing. • Designed an agent execution engine supporting ReAct and Plan-and-Execute strategies. • Added multi-agent collaboration and parallel tool calls with ordered result returns. • Built browser automation security via Chrome DevTools Protocol integration and HITL approval.