Multi-agent quorum workflow — fans out tasks to Claude, Codex, Copilot, Cursor, and Gemini in parallel, then reconciles their outputs into a high-confidence recommendation. Use for planning, code review, implementation validation, bug diagnosis, and architecture decisions.
Multi-agent MCP server that fans out prompts to Claude, Codex, Copilot, Cursor, and Gemini in parallel, then reconciles their outputs into a single high-confidence recommendation.
Then install the agent CLIs you want to use (see Prerequisites), and run:
/quorum:run review this repo
Two usage paths
Quorum works in two distinct ways depending on your editor:
Claude Code — full orchestration
Claude Code gets the complete experience: skills, relay agents, and blind judge reconciliation.
/quorum:run <prompt> — Fan out to all enabled agents, deduplicate findings, run blind judge scoring, and present a synthesized recommendation. Supports modes: review, review-codebase, architecture, implement, diagnose.
/quorum:agent <agent> <prompt> — Test a single agent and get the raw result back. Useful for debugging.
Claude participates as both an agent (analyzing code directly) and the judge (scoring anonymized outputs). Agent identities are hidden during scoring to prevent bias.
Any MCP client — direct tool calls
Cursor, VS Code, Windsurf, and any other MCP-compatible client get access to the MCP tools directly. The client's own LLM decides how to use the results.
codex_query / copilot_query / cursor_query / gemini_query — Query a single agent
quorum_query — Fan out to all enabled agents in parallel and return collected results
No skills, no relay agents, no blind judge — your editor's LLM handles interpretation.
nexusentis