Skip to content
Conduit

mcp server

pingfusi

MCP server + CLI that puts a real human in your coding agent's loop. It publishes work mid-task, a reviewer pins what's wrong and returns a verdict, and the agent iterates until approved.

alex-durango/pingfusicommunitylistedunreviewednot running here yet
Use tool

Set up

Runs on your computer. Pick your agent and operating system.

Before you start

  • Node.js 20 or newer
~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "pingfusi": {
      "command": "npx",
      "args": [
        "-y",
        "pingfusi"
      ]
    }
  }
}

same format Cursor, GitHub Copilot CLI, Junie, JetBrains AI Assistant, Augment (Auggie CLI), Factory Droid, Kiro, Devin CLI +12

Values in angle brackets are placeholders. Restart Claude Desktop after editing its config.Claude Desktop docs

README

synced Oct 15, 2026

About

pingfusi is an MCP that lets AI agents call human reviewers.

Think MTurk for AI agents. With pingfusi your agent can:

  • Get a second opinion: a real person looks at the work, not another model
  • Get human judgment: where opinion is the answer — design taste, wording, which version people prefer
  • Skip the iteration loop: the agent revises, another human checks each round, and you only see the finished version

Quickstart

Set up Pingfusi for your coding agents with a single command.

npx pingfusi setup

Example Prompts

Here are some example prompts you can try with the pingfusi MCP.

feedback aboutexample promptwhat you getdemo
a confusing pageIs my pricing page confusing anywhere? use pingfusicomments pinned to what's off
a website cloneClone www.example.com pixel-perfect. use pingfusia perfectly cloned websitecopy-anything.com
design tasteMake my website not look like AI slop. use pingfusidesign feedbackvideo
video vibesDoes my promo video look right? use pingfusifeedback pinned to timestamps

Each of these publishes the work first so the reviewer can open it; when in doubt, publish: a reviewer can only judge what they can see. Text-only questions work too — if words can carry the whole question, the reviewer just sees the question.

Credits

Two lifetime reviews are free. The paid plan includes 100 Pingfusi credits for $30/month, with no rollover. Game playtesting is available through Qaping, with its own credits; it is not a Pingfusi review job.

How it works

Every job is the same loop underneath:

  1. File a review — the agent pushes the work (a built site, a video, any artifact) so the reviewer can open it, then files a review with concrete steps to check and what counts as approval.
  2. A human reviews — a real person opens the work and sends back pinned comments anchored to the exact elements that are off.
  3. You get better output — the reviewer's comments land in the agent's context, so the next version is shaped by real human feedback.

Example: you ask for a landing page that doesn't look like AI slop. The agent publishes it, a reviewer answers "gradient looks template-y" and "too much padding under the hero". The agent fixes both and refiles; the next reviewer approves. You come back to a page a human signed off on.

CLI commands

The full command lives in docs/COMMANDS.md.

pingfusi setup                          install + onboarding
pingfusi doctor                         check the install; prints a fix per problem
pingfusi ask "<question>"               ask a human reviewer, from any directory
pingfusi studio [ping_id]               local viewer for review results —
                                        sessions, recordings, transcripts, key moments,
                                        agent findings (contract: docs/STUDIO.md)

License

MIT

↑↓ navigate openesc close