Show HN: Roundtable MCP Server to Use Claude, Cursor, Codex, Gemini from One UI (github.com)

Roundtable AI is a local Model Context Protocol (MCP) server that lets a primary IDE assistant delegate parts of a single prompt to multiple specialized sub-agents (Gemini, Claude, Codex, Cursor, etc.), run them in parallel, and synthesize a unified response. It preserves a shared "context bundle" across sub-agents so everyone sees the same project state, removes manual copy‑paste between tools, and leverages each model’s strengths (e.g., Gemini’s 1M‑token context for large-code analysis, Claude for reasoning, Codex for implementation). The tool integrates with 26+ MCP-compatible clients (VS Code, JetBrains, Cursor, Claude Code), invokes the CLIs you already have configured (using CLI_MCP_SUBAGENTS env var), and claims no markup or added API costs—just your existing subscriptions. For engineers, this enables workflows like multi‑stack incident war rooms or performance tuning where frontend analysis, backend fixes, and infra diagnostics are parallelized and then aggregated into a single incident report or optimization plan. Technically, Roundtable dispatches tasks to local CLI sub-agents over stdio, collects their outputs, and synthesizes results back to the primary assistant. Quick start: pip install roundtable-ai and run roundtable-ai --agents codex,claude (or add via client MCP configs). The approach reduces wait time, enforces deterministic multi-agent orchestration, and simplifies end‑to‑end debugging and code-rewrite pipelines by combining model specialization with shared context.