2ndopinion-cli

  ●━━●
  └━━●  2ndOpinion

The first platform where Claude, Codex, and Gemini communicate with each other to review your code. 10+ AI-powered tools via MCP, REST API, and CLI.

---

Three Ways to Use

MCP Server (for Claude Code, Cursor, etc.)

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "2ndopinion": {
      "command": "npx",
      "args": ["-y", "2ndopinion-cli", "mcp"]
    }
  }
}

Or with an API key (no CLI login required):

{
  "mcpServers": {
    "2ndopinion": {
      "command": "npx",
      "args": ["-y", "2ndopinion-cli", "mcp"],
      "env": {
        "SECONDOPINION_API_KEY": "sk_2op_your_key_here"
      }
    }
  }
}

Get your API key at get2ndopinion.dev/dashboard/keys.

All tools appear natively in your AI assistant's context.

CLI

npm install -g 2ndopinion-cli

2ndopinion signup          # Create an account
2ndopinion                 # Analyze current git changes
2ndopinion fix             # Auto-fix detected risks
2ndopinion watch           # Continuous monitoring

REST API

curl -X POST https://get2ndopinion.dev/api/gateway/opinion \
  -H "X-2ndOpinion-Key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"diff": "...", "llm": "codex"}'

Generate API keys with 2ndopinion keys create.

---

Tools

Every tool works across MCP, API, and CLI.

---

What Makes It Different

Confidence-Weighted Consensus — When 3 models review your code, they're not weighted equally. 2ndOpinion tracks which model is most accurate for each language and issue type, then weights the consensus by proven performance.

Smart Model Routing — Use --llm auto and the system picks the best model for your language based on real accuracy data. No guessing.

Pattern Memory — Known bug patterns are flagged instantly before the LLMs even run. When a pattern has been seen hundreds of times across the platform, you get an instant heads-up.

Regression Tracking — See what changed since your last analysis: "2 new issues introduced, 1 resolved." Your review has memory.

Project Context — Save your project's framework, dependencies, and structure once. Every subsequent analysis is tailored to YOUR codebase.

Custom Rules — Define rules in plain language ("Always flag eval()") and they're enforced on every analysis. Team rules are shared across all members.

---

Skills Marketplace

Build and sell custom AI tools. Encode your expertise as a reusable skill, set a price (1–20 credits per run), and earn 70% of every execution.

Browse and run skills at get2ndopinion.dev/marketplace.

---

Developer Platform

Build on the 2ndOpinion API with official SDKs, webhooks, batch processing, and streaming.

SDKs:

# JavaScript / TypeScript
npm install @2ndopinion/sdk

# Python
pip install secondopinion-sdk

Webhooks — Real-time events (analysis.completed, analysis.high_risk, credits.low, skill.run.completed), HMAC-SHA256 signed.

Batch — Analyze up to 20 files per request via POST /api/gateway/batch.

Streaming — Real-time SSE results via POST /api/gateway/opinion/stream, /consensus/stream, /debate/stream.

Build a Skill:

curl -X POST https://get2ndopinion.dev/api/skills \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "My Custom Audit",
    "slug": "my-custom-audit",
    "description": "Checks for common issues in Express.js apps",
    "category": "security",
    "creditCost": 3,
    "executionMode": "consensus",
    "systemPrompt": "You are a security auditor specializing in Express.js..."
  }'

Full developer docs at get2ndopinion.dev/docs/api.

---

CLI Commands

---

How It Works

1. AI writes code — Write code however you like, with any AI assistant or by hand
2. Three AIs cross-check it — Claude, Codex, and Gemini independently review the same diff. 2ndOpinion weights their findings by proven accuracy, checks against known bug patterns, and shows where they agree and disagree
3. Ship with confidence — Get accept/review/reject verdicts, auto-fix issues, generate tests, or run a security audit before merging

---

Pricing

Individual:

Teams:

Teams share a credit pool. Every member's analysis makes the team's reviews smarter through collective pattern memory and shared rules.

Need more? Buy credit packs anytime: Starter (100 credits / $5), Builder (500 / $20), Scale (2,500 / $100).

All plans include access to Claude Sonnet 4, Codex, and Gemini 1.5 Pro via MCP, API, and CLI. Tools cost 1–7 credits depending on complexity.

---

FAQ

Is my code stored? No. Diffs are sent to the chosen LLM for analysis and discarded. We never store your source code.

Which LLMs are supported? Claude Sonnet 4 (Anthropic), Codex (OpenAI), and Gemini 1.5 Pro (Google).

Does it work with any language? Yes. If git can diff it, 2ndOpinion can review it.

Can I use it with Claude Code / Cursor / Copilot? Yes — add the MCP server config above, or run 2ndopinion watch in a separate terminal.

What's the difference between single-model and consensus? Single-model (1 credit) sends your diff to one LLM. Consensus (3 credits) sends it to all three and shows where they agree and disagree — with confidence weighting based on each model's historical accuracy.

What does "auto" do for --llm? It picks the best model for your language based on real performance data from thousands of analyses. If we don't have enough data for your language yet, it defaults to Claude.

Do teams share credits? Yes. Team plans have a shared credit pool. Any member can use credits and it deducts from the team total. You can see per-member usage in the team dashboard.

---

---

CI/CD & Scripting

Use --ci (or --json) to get structured JSON output with no interactive UI. Use --plain for human-readable text output without ANSI formatting.

Flags

Exit Codes

Examples

# JSON analysis
2ndopinion analyze --ci
2ndopinion analyze --ci --llm claude
2ndopinion status --ci

# Parse with jq
2ndopinion analyze --ci | jq '.verdict'
2ndopinion status --ci | jq '.credits.remaining'

# Plain text (no ANSI)
2ndopinion analyze --plain
2ndopinion status --plain

# Use exit code in scripts
2ndopinion analyze --ci
if [ $? -eq 2 ]; then echo "Code rejected!"; fi

GitHub Actions

- name: AI Code Review
  run: |
    npx 2ndopinion analyze --ci --llm codex > review.json
    VERDICT=$(jq -r '.verdict' review.json)
    if [ "$VERDICT" = "reject" ]; then
      echo "::error::Code review failed"
      exit 1
    fi
  env:
    SECONDOPINION_API_KEY: ${{ secrets.SECONDOPINION_API_KEY }}

GitLab CI

code-review:
  script:
    - npx 2ndopinion analyze --ci --llm codex > review.json
    - cat review.json | jq '.summary'
  artifacts:
    paths:
      - review.json

JSON Output Format

analyze --ci:

{
  "verdict": "review",
  "model": "codex",
  "findings": [
    {
      "severity": "HIGH",
      "category": "security",
      "title": "SQL injection via string interpolation",
      "description": "The query builds SQL unsafely",
      "file": "auth.js",
      "line": 9,
      "suggestion": "Use parameterized queries"
    }
  ],
  "summary": "Found 1 issue: 1 HIGH",
  "creditsUsed": 1,
  "creditsRemaining": 4
}

status --ci:

{
  "tier": "free",
  "email": "dev@example.com",
  "credits": { "used": 3, "limit": 5, "remaining": 2, "bonus": 0 },
  "defaultLlm": "codex",
  "resetsAt": "2026-05-01T00:00:00.000Z"
}

---

Full docs at get2ndopinion.dev

License

Proprietary. All rights reserved.