MCP 服务器

Cynical Sally

Provides brutally honest code reviews, explanations, and refactoring suggestions via CLI and MCP tools.

README

<h1 align="center">Cynical Sally</h1>

Brutally honest code reviews. In your terminal and your AI editor. Because "You're absolutely right" is probably absolutely wrong. <code>npm i -g @cynicalsally/cli</code> · MCP server for Claude Code, Cursor & Windsurf

Your AI pair programmer is lying to you. Sally isn't.

She's the senior engineer your code hoped it'd never meet. Scores from 0 to 10, real issues backed by evidence, and fixes you can actually use.

Works as a CLI tool and as an MCP server in Claude Code, Cursor, and Windsurf.

Install

npm install -g @cynicalsally/cli

Or run without installing:

npx @cynicalsally/cli roast ./src/

Requirements: Node.js 18+

See exactly what leaves your machine

Sending code to a server you don't control deserves more than "trust me." So Sally lets you verify it instead:

sally roast --dry-run ./src/

--dry-run collects everything as if it were about to roast — then sends nothing. Instead it prints the exact payload: every file path, byte size, and token estimate; which files were held back and why (.env, keys, certs, binaries, .gitignore matches, size limits); and writes a local SHA-256 receipt to .sally/ so you can verify byte-for-byte what would have been uploaded.

WOULD SEND 3 files · 91 B · ~24 tokens (est.)
  src/app.ts        36 B   ~9 tok   68a3be428746…

HELD BACK 2 items — kept on your machine
  ✖ secret (2) — looks like a secret — never leaves your machine
      .env
      server.key

Secret files (.env, SSH keys, certs, credential files) are skipped on your machine before anything is sent — verify it yourself with --dry-run. Only review code you're allowed to upload. Local reports land in .sally/ — add it to your .gitignore:

.sally/

See Privacy & Security below for the full data-flow.

Quick Start

# Sally auto-detects what to review
sally roast
# → staged changes? reviews those
# → unstaged changes? reviews those
# → recent commit? reviews that
# → nothing? scans the directory

# Roast a file or directory
sally roast src/utils/auth.ts
sally roast ./src/

# Roast staged changes before you commit
sally roast --staged

# Compare your branch against main
sally roast --diff main

# Deep analysis with issues + actionable fixes
sally roast ./src/ -m full_truth

# Run deep analysis in the background (OS notification when done)
sally roast ./src/ -m full_truth --bg

# See exactly what would be sent — and send nothing
sally roast --dry-run ./src/

# Get a shareable roast card (saved to .sally/)
sally roast ./src/ --card

Roast Options

sally roast [paths...] [options]

  --staged              Review only staged git changes
  --diff <branch>       Compare against another branch (e.g., main)
  -m, --mode <mode>     "quick" (default) or "full_truth" (deep dive)
  --tone <tone>         "cynical" (default), "neutral", or "professional"
  --lang <lang>         Response language code (default: "en")
  --json                Output raw JSON (for piping or scripting)
  --fail-under <score>  Exit code 1 if quality score is below threshold
  --ci                  CI mode: compact output, exit codes
  --bg                  Run Full Truth in background, get OS notification when done
  --dry-run             Print the exact payload (files, sizes, tokens, SHA-256) and send NOTHING
  --card                Print + save a shareable roast card after the review

<h2 align="center">Sally's Full Suite</h2>

6 tools. Unlimited usage. The most honest code reviewer you'll ever work with — in your terminal and your AI editor.

Explain

Sally reads the spaghetti someone left in your codebase and translates it into plain English. Just the cold, clear truth of what it actually does.

sally explain src/utils/auth.ts

# Pipe code directly
cat legacy-module.js | sally explain

# Explain the current directory
sally explain

Refactor

Before and after, side by side. Sally explains why one of them is going to haunt your 3am on-call rotation.

sally refactor src/components/Dashboard.tsx

# Refactor current directory
sally refactor

PR Review

Sally reviews your PR like a senior engineer who has time, opinions, and absolutely no reason to be polite.

# Review PR #42 (requires GitHub CLI)
sally review-pr 42

# Review current branch vs main
sally review-pr

# Pipe a diff
git diff main | sally review-pr

Brainstorm

Pitch your architecture idea and Sally tells you the three ways it falls apart at scale. Cheaper than a post-mortem.

sally brainstorm "Microservices for a 2-person team?"

# Brainstorm about the current project
sally brainstorm

Frontend Review

Sally tells you why your component re-renders on every keystroke and why your z-index is load-bearing.

sally frontend src/components/Header.tsx

# Review all frontend code in a directory
sally frontend ./src/

Marketing Review

Run your copy by Sally before your customers do. They won't be this constructive about it.

sally marketing "Ship faster with AI-powered code reviews"

# Review your README and landing page copy
sally marketing README.md

Every tool accepts file paths, raw text, or piped stdin. Each includes 1 free trial, no account needed.

CI/CD Integration

Gate your pipeline on code quality:

# GitHub Actions
- name: Sally Code Review
  run: npx @cynicalsally/cli roast ./src/ --fail-under=5 --ci

--ci gives compact output with exit codes. --fail-under fails the build when the score drops below your threshold. Add --json for machine-readable output.

MCP Server

Sally works as an MCP server inside Claude Code, Cursor, and Windsurf.

Claude Code

claude mcp add cynical-sally -- npx @cynicalsally/cli mcp

Cursor

Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

{
  "mcpServers": {
    "cynical-sally": {
      "command": "npx",
      "args": ["@cynicalsally/cli", "mcp"]
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "cynical-sally": {
      "command": "npx",
      "args": ["@cynicalsally/cli", "mcp"]
    }
  }
}

Available tools

MCP Tool	What it does
`sally_roast`	Code review with score, issues, and fixes
`sally_explain`	Explain code with Sally's personality
`sally_review_pr`	Review PR diffs
`sally_refactor`	Refactoring suggestions with before/after
`sally_brainstorm`	Feedback on ideas and approaches
`sally_frontend`	Frontend/UI code review
`sally_marketing`	Marketing copy review
`sally_usage`	Check quota and account status

Roast by path — the agent can call sally_roast with just paths (files or directories); Sally reads them locally and skips binaries and secret files, so the agent doesn't have to read and pass content itself.

Prompts — Sally also exposes ready-made slash-command intents (roast, review-pr, explain) in clients that surface MCP prompts.

Run sally mcp in your terminal to see setup instructions.

All Commands

Command	Description
`sally roast [paths...]`	Review files, directories, or git changes
`sally explain [file]`	Explain what code actually does
`sally refactor [file]`	Refactoring with before/after code
`sally review-pr [pr]`	Review a PR diff
`sally brainstorm "idea"`	Feedback on ideas and approaches
`sally frontend [file]`	Frontend/UI code review
`sally marketing "copy"`	Marketing copy review
`sally login <email>`	Log in via magic link
`sally logout`	Clear stored session
`sally usage`	Check your quota and account status
`sally upgrade`	Upgrade to Sally's Full Suite
`sally results`	View background review results
`sally mcp`	MCP server setup instructions

Free to Use

90 free roasts per month, no account needed. Every premium tool includes a free trial.

sally usage     # Check your quota
sally upgrade   # Unlock the Full Suite

Privacy & Security

Your code is yours. Don't take our word for it — run sally roast --dry-run and see the exact payload before anything is sent. Here's what happens to it:

Verify before you send. --dry-run prints every file, size, token estimate, and a SHA-256 receipt of exactly what would be uploaded — and sends nothing. The MCP sally_roast tool has the same preview mode.
Sent only to be reviewed. The files you choose are transmitted over HTTPS and processed in real-time to generate the review — that's the only reason they leave your machine.
Never written to disk, logs, or analytics. Your source code is processed in memory and discarded after analysis. It is never persisted to a database, never written to application logs or error traces, and never sent to any third-party APM or analytics. We keep the review (score, issues), not your source.
Never trained on, sold, or shared. Analysis runs through Anthropic's API, which doesn't train on submitted content.
Only what you point at. Sally doesn't browse your repo, read files you didn't give her, or scan your projects or plans. Secret files (.env, keys, certs, credential files) are skipped on your machine before anything is sent — and --dry-run shows you exactly which ones.
Anonymous by default. Reviews are tied to a random device ID, not your identity — until you link an email for Full Suite. Config stored locally at ~/.sally/config.json.
Signed releases. npm packages are published with provenance — a cryptographic, public attestation linking each release to the exact source commit and CI build that produced it.

Full engineering detail — data-flow diagram, what's retained, subprocessors, and log policy — is in docs/PRIVACY.md. User-facing summary: cynicalsally.com/privacy.

Contributing

Found a bug or have a feature idea? Open an issue. Sally promises to only judge your issue title a little.

License

MIT

<a href="https://cynicalsally.com">cynicalsally.com</a> · <a href="https://www.npmjs.com/package/@cynicalsally/cli">npm</a> · <a href="https://github.com/w1ckedxt/cynical-sally/issues">issues</a>