Merv was created to solve two problems: agent reliability, and the visibility of agent work.
Models and harnesses get better at the task every month. So Merv is not an agent harness. It is a gating mechanism: agent work moves forward only by passing adversarial review.
claude code · codex · cursor · gemini cli · opencode
The plan is reviewed before anything runs. The report is reviewed before anything counts. Rejections send the work back.
After each wave of experiments, five lenses ask what worked, what failed, and what hasn’t been tried. The synthesis is reviewed and sets up the next wave.
Adapters plug into the agents you already use. The backend holds the research state. The frontend lets you watch — from strategy to a single run. Execution goes to cloud sandboxes or your own machine; your checkout and SSH keys never leave it.
The code is public — the gates, the reviewer prompts, the proxy that touches your files. A gate you can read.
claude plugin marketplace add https://rapidreview.io/marketplace.json claude plugin install merv@rapidreview
Restart Claude Code and connect a project. Codex, Cursor, Gemini CLI, OpenCode: setup guide.