What it does

Roundtable AI runs on your own machine as a Model Context Protocol (MCP) server, for people who use more than one AI coding assistant. From a single prompt in an MCP-compatible IDE, your main assistant can delegate parts of a job to sub-agents running Gemini, Claude Code, Codex or Cursor, and receive one combined answer back. The project's README says the main assistant passes shared project context to every sub-agent, that the sub-agents run in parallel, and that Roundtable calls the CLI tools already installed on your machine using your existing API keys and subscriptions, with no markup added. You install it with pip or uvx, run roundtable-ai --check to see which assistants are available, and choose which ones to enable with the --agents option or the CLI_MCP_SUBAGENTS environment variable. The README lists setup instructions for more than 26 MCP-compatible clients, including Claude Code, Cursor, VS Code, JetBrains IDEs, Claude Desktop, Windsurf and Gemini CLI. It exposes MCP tools to check each assistant's availability and to run a task through it. The project is licensed under the MIT License.

Capabilities

Delegates tasks from one IDE prompt to Gemini, Claude Code, Codex and Cursor sub-agents
Runs sub-agents in parallel and combines their responses
Shares project context across all sub-agents
Uses CLI tools and subscriptions you already have, with no markup
Setup instructions for 26+ MCP-compatible clients
Install with pip or uvx and check availability with a single command

When to use it

Splitting a bug that spans frontend, backend and infrastructure across several assistants
Combining the strengths of different coding assistants without copying between chats
Adding multi-assistant delegation to an IDE you already use