roundtable

Roundtable AI MCP Server is a local multi-agent coordination hub. It orchestrates specialized sub-agents (Gemini, Claude, Codex, Cursor, etc.) to solve engineering problems by running tasks in parallel, sharing context, and synthesizing a single, cohesive response. Start the server from your environment and interact with it using the MCP client tooling or IDE integrations that support the MCP protocol. Once running, you can dispatch prompts that delegate parts of a task to multiple sub-agents and receive a unified result that combines analysis, reasoning, and implementation suggestions from the respective tools. The server is designed to work with your existing CLI tools and API subscriptions, avoiding any vendor-specific markup.

You can use Roundtable AI by starting the MCP server and then issuing commands to enable all available agents or select specific ones (for example: gemini, claude, codex, cursor). The workflow typically involves packaging a prompt and context, routing subtasks to sub-agents in parallel, and then aggregating their outputs into a final synthesized answer. IDE integrations and prompt templates provided with the server help you craft multi-agent prompts and define how results should be presented in your IDE or chat interface.

Prerequisites:

• Python 3.10+ installed on your machine
• pip available in your environment
• Optional: a local environment with your preferred IDE (VS Code, JetBrains, Cursor, etc.)

Installation steps:

1. Create and activate a Python virtual environment (recommended): python -m venv .venv source .venv/bin/activate # macOS/Linux .venv\Scripts\activate # Windows

2. Install the Roundtable MCP server package from PyPI: pip install roundtable-ai

3. Run the MCP server (as defined in mcp_config): python -m roundtable_ai

4. Verify the server is running and listening on the default port (usually localhost:8080 or configured by the tool you use).

5. If your IDE integration requires a configuration file, use the provided defaults or your environment overrides as needed.

Tips and caveats:

• Ensure your Python environment has network access if you’re using remote sub-agents or external APIs.
• If you encounter agent-specific errors, check that the CLI tools for Gemini, Claude, Codex, Cursor, etc., are installed and accessible in your PATH.
• Some installations may require authentication tokens or API keys for your preferred models; keep these credentials secure and load them via environment variables or your IDE’s secret manager.
• You can customize the set of agents via your MCP client by specifying --agents or equivalent options supported by your tooling configuration.
• If running behind a firewall, ensure outbound connections to the sub-agent services are allowed or configure local proxies as needed.