consult: open-source MCP server lets Claude Code, Codex and Grok give each other second opinions
Equivalent_Soup4149 · reddit · 2026-10-21
The author open-sourced consult (MIT), an MCP server that lets coding agents like Claude Code, Codex and Grok request second opinions from each other. It runs through the agent CLIs you're already signed into — no API keys — and is listed on the Official MCP Registry (io.github.neelsatyavolu/consult, v1.2.0).
Core tools:
- askagent / askagents: get advice from one or several other agents
- reviewchanges: have installed agents review the current diff; --base main covers a whole branch, and v1.2.0 adds --pr 42 to review a GitHub PR as GitHub renders it
- Optional: listsessions, asksession (query a read-only fork of a live session), sendmessage (leave a note for a running session and wake idle Claude Code/Codex sessions)
You can also skip MCP: npx -y consult-mcp@latest review.
The author raises three design trade-offs for feedback: advisors run headless and read-only with no MCP servers loaded (prevents recursion but limits their tooling); the server ships instructions on when to ask for help; and sendmessage notes are marked as peer-agent, not user, to avoid instruction injection. Works on Mac/Linux/Windows; messaging and live sessions are Mac/Linux only.
More from coding & agent
- Burkov: without explicit checklists, coding agents hand you 80%-done work — burkov · 2026-10-21
- Microsoft's Self-HPO: Coding Agent Redesigns Its Own Harness, +7 Points on SWE-bench — mark_k · 2026-10-21
- Laude Institute launches Slingshots FOUR: 14 projects across 8 top institutions — a1zhang · 2026-10-21
- Research agent Épi sets 120 records, narrows n-queens constant interval 84x — dair_ai · 2026-10-21
- Compaction hooks in practice: keep last n messages verbatim, summarize the rest for full agent continuity — MoonL88537 · 2026-10-21
- The blank prompt box hides real operational burden: context, workflow and supervision fall on users — _AustinCalvert_ · 2026-10-21