CLI and MCP for model fusion via harnesses

Name: CLI and MCP for model fusion via harnesses
Availability: InStock
Author: kerryritter

by kerryritter·Jun 24, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidBig BrainNiche Gem

Local multi-agent fusion beats single-model blind spots without new API keys.

Strengths

•Zero dependencies Rust binary drives existing agent CLIs, no vendor lock-in or new auth.
•Dual CLI and MCP surfaces let humans or agents convene the panel interchangeably.
•Consensus flags highlight disagreements worth attention, not just merged outputs.

Weaknesses

•OpenRouter Fusion already offers multi-model deliberation as a hosted service.
•Running multiple agents locally adds latency and cost compared to single-model calls.

Post Description

After seeing OpenRouter Fusion and Sakana's results, I decided to take some learning from my building my start-up's harness and extract a CLI/MCP for agent-to-agent communication. Simple things like "Use parley to have kimi review this work {from Claude}" continually finds notable gaps. The jump in quality has been huge.