cli

The CLI for Nullify — a fully autonomous AI workforce for product security.

Scan, triage, fix, and track security vulnerabilities across your entire stack — from code to cloud — directly from your terminal. The Nullify CLI connects to your Nullify instance and gives you access to SAST, SCA, secrets detection, DAST/pentesting, bug hunting, CSPM, AI-powered autofix, and interactive AI security agents.

Features

• Unified findings — Query vulnerabilities across all scanner types (SAST, SCA, secrets, DAST, CSPM) in a single command
• AI-powered remediation — Generate fix patches and open PRs automatically
• CI/CD quality gates — Block deployments when critical findings are present
• DAST/Pentest scanning — Run API security scans in the cloud or locally via Docker
• MCP server — 50+ tools for AI coding assistants (Claude Code, Cursor, VS Code, and more)
• Multi-instance support — Manage multiple Nullify instances and switch between them

Installation

macOS / Linux

curl -sSfL https://raw.githubusercontent.com/Nullify-Platform/cli/main/install.sh | sh

Pre-configure your instance during install:

curl -sSfL https://raw.githubusercontent.com/Nullify-Platform/cli/main/install.sh | sh -s -- --host api.acme.nullify.ai

Go

go install github.com/nullify-platform/cli/cmd/cli@latest

Windows

Download the latest .zip for your architecture from GitHub Releases and add the binary to your PATH.

Verify

nullify --version

The installer verifies SHA-256 checksums automatically. Binaries are available for Linux, macOS, and Windows on both amd64 and arm64.

Quick Start

# Set up your instance, authenticate, and configure MCP — all in one step
nullify init

# Check your security posture
nullify status

# View findings across all scanner types
nullify findings

Authentication

Interactive (SSO/IdP)

nullify auth login

Opens your browser to sign in with your identity provider (for example Microsoft Entra ID or GitLab). Tokens are stored locally and refreshed automatically.

With no host saved, the CLI signs in through nullify.ai, the Nullify router. With a host saved or passed, it first asks that instance where its login runs (GET /auth/cli/config): an instance on the router hands the login to the router, and an instance that signs in itself keeps its own login. The CLI only follows an answer naming your instance or a Nullify router, and fails otherwise.

# Through the router, or wherever your saved instance says
nullify auth login

# Name your instance; on the router this still signs in at the router
nullify auth login --host acme.nullify.ai

# Choose the organization to start with
nullify auth login --org acme-web

One router login covers every instance the router lists for you, so --host <instance> works afterwards without another login, and an older login an instance ran itself is replaced. After signing in, the CLI selects the organization commands run as: the one you pass with --org, the one you used last, the only one, or the one you pick in a terminal.

Organizations

One login can cover several organizations, on one instance or several. Each command runs as the first of these that applies:

1. --org <name|id>
2. NULLIFY_ORG (name or id)
3. NULLIFY_GITHUB_OWNER_ID
4. The owner of the current directory's git remote, if it names exactly one of your organizations on the instance (worktrees and submodules included)
5. Your default, set by nullify org use or the last login
6. Your only organization on the instance

If your default is no longer available, the command stops and says so rather than running as another organization; in a terminal it asks. When nothing applies and you have several organizations, the CLI asks in a terminal and remembers the answer; elsewhere it fails and lists the choices. nullify org list shows your organizations and which rule picked the one in use.

Signed in through the router, nullify org use can name an organization on another instance: plain commands then run on that instance. --org does the same for one command when the command's host is the router (nullify.ai, the default after a router login).

# See your organizations on this instance and which one is in use
nullify org list

# Make one your default
nullify org use acme-web

# Run one command as another
nullify findings --org acme-payments

An API token (NULLIFY_TOKEN) belongs to one organization, and the server uses it when the command names none. With a token, only --org, NULLIFY_ORG and NULLIFY_GITHUB_OWNER_ID apply (the first two are looked up on the instance itself); nullify org use, the git remote and the prompt are for browser logins. An API token always needs your instance as the host (--host or NULLIFY_HOST), never nullify.ai. When NULLIFY_TOKEN is set it wins over your browser login, and nullify auth status says so.

Upgrading from an earlier version

Your login and your selected organization carry over. If you used an API token as a workaround for the router login, remove NULLIFY_TOKEN and NULLIFY_GITHUB_OWNER_ID from your shell and your MCP configuration, run nullify auth login, and restart your MCP servers. Upgrade every copy of the CLI your tools run: an older copy running alongside can overwrite the login file.

Environment Variables

export NULLIFY_HOST=api.acme.nullify.ai
export NULLIFY_TOKEN=your-token-here
export NULLIFY_ORG=acme-web   # optional: organization to run as, by name or id

Multiple Instances

# Log in to a second instance
nullify auth login --host api.staging.nullify.ai

# Switch the active instance
nullify auth switch --host api.staging.nullify.ai

# List configured instances
nullify auth switch

Commands

Core

Authentication

Scanning

CI/CD

Tooling

Findings

# All findings across every scanner
nullify findings

# Filter by severity and scanner type
nullify findings --severity critical --type sast

# Filter by repository and status
nullify findings --repo my-repo --status open

# Limit results
nullify findings --limit 50

# List CSPM or SCPM findings in one UTC window
nullify findings list --type cspm --from 2026-11-01T00:00:00Z --to 2026-11-02T00:00:00Z
nullify findings list --type scpm

--type accepts sast, sca_dependencies, sca_containers, secrets, secrets_sensitive_data, pentest, bughunt, cspm and scpm. An unknown type is an error.

--type secrets lists credentials and sensitive data together. To act on a listed finding, pass the type value of its row, for example nullify findings get abc123 --type SecretsSensitiveData. In a verb, secrets means credentials only.

The output reports total, returned and truncated:

Act on a finding. Allowlisting needs a reason and an explicit type: UserFalsePositive when triage was wrong, UserAssumeRisk when the risk is real and accepted, UserFixed or UserOther otherwise. The server may not record the type for bughunt, sca_dependencies, sca_containers and secrets_sensitive_data yet. For these types the command writes a warning to stderr and adds a warning field to the result. The reason is recorded. For a CSPM false positive, use findings mark-false-positive: allowlisting with UserFalsePositive does not set is_false_positive on a CSPM finding.

nullify findings mark-false-positive abc123 --type cspm --reason "Bucket is public by design"
nullify findings allowlist abc123 --type sast --allowlist-type UserFalsePositive --reason "Test fixture"
nullify findings unallowlist abc123 --type cspm --reason "Bucket is no longer public"
nullify findings archive abc123 --type cspm
nullify findings feedback abc123 --verdict wrong --comment "Not reachable from the internet"

Raw API commands follow the API spec: nullify api cspm create-allowlist-by-id, create-unallowlist-by-id, create-archive-by-id, create-unarchive-by-id, create-mark-false-positive-by-id, create-unmark-false-positive-by-id, patch-findings, get-triage and delete-ticket.

Command stability

spec/command-names.lock fixes the name of each API command, and spec/command-surface.tsv lists every command and flag. CI fails if a command or flag disappears unless internal/commands/surface_decisions.go records the decision. A moved command keeps working as a deprecated alias that calls the new route. See CHANGELOG.md for the moves.

Pentest Scans

Cloud Scan

nullify pentest \
  --app-name      "My REST API" \
  --spec-path     ./openapi.json \
  --target-host   https://api.myapp.dev \
  --github-owner  my-org \
  --github-repo   my-repo \
  --header        "Authorization: Bearer token123"

Local Scan (Docker)

For APIs that aren't publicly accessible. Requires Docker.

nullify pentest \
  --app-name      "My REST API" \
  --spec-path     ./openapi.json \
  --target-host   http://localhost:8080 \
  --github-owner  my-org \
  --github-repo   my-repo \
  --local

Add --use-host-network if the target is running directly on the host machine.

CI/CD Integration

Quality Gate

Block PRs and deployments when findings exceed a severity threshold:

# .github/workflows/security.yml
name: Security Gate
on: [pull_request]

jobs:
  security:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Install Nullify CLI
        run: curl -sSfL https://raw.githubusercontent.com/Nullify-Platform/cli/main/install.sh | sh

      - name: Security Gate
        run: nullify ci gate --severity-threshold high
        env:
          NULLIFY_HOST: ${{ vars.NULLIFY_HOST }}
          NULLIFY_TOKEN: ${{ secrets.NULLIFY_TOKEN }}

PR Security Summary

Add a security report to your pull request:

      - name: Security Report
        run: nullify ci report >> $GITHUB_STEP_SUMMARY
        env:
          NULLIFY_HOST: ${{ vars.NULLIFY_HOST }}
          NULLIFY_TOKEN: ${{ secrets.NULLIFY_TOKEN }}

MCP Server

The CLI includes a built-in Model Context Protocol server with 50+ tools, giving AI coding assistants full access to your security data. Capabilities include:

• Query — findings, repositories, SBOMs, cloud accounts, metrics, trends
• Triage — allowlist, archive or mark false positive across every finding type, including CSPM, reopen, and record feedback
• Remediate — start AI-powered fixes, then follow them with nullify_get_fix_status and nullify_get_fix_diff
• Track — campaigns, action items, SLA policies, code reviews

nullify mcp serve --tools findings-write adds the tools that the default set leaves out: archive, unarchive, mark and unmark false positive, and delete ticket. --tools all also has them. Every tool that changes state carries the destructive annotation, so a client can ask a person first. spec/mcp-tools.tsv lists every tool, its access and its tool sets.

Setup

Add the following to your MCP config file:

{
  "mcpServers": {
    "nullify": {
      "command": "nullify",
      "args": ["mcp", "serve"]
    }
  }
}

One server serves one instance. When your login covers several organizations there, every tool takes an optional org argument (name or id), and the list_organizations tool lists them; every result says which organization it ran as. Without the argument a call uses the default the server settled when it started (--org, then the same order as the CLI); nullify org use in a terminal does not change a running server. If that leaves several organizations, the call returns the choices instead of guessing. To give each organization its own server, add one entry per organization:

{
  "mcpServers": {
    "nullify-acme-web": {
      "command": "nullify",
      "args": ["mcp", "serve", "--org", "acme-web"]
    },
    "nullify-acme-payments": {
      "command": "nullify",
      "args": ["mcp", "serve", "--org", "acme-payments"]
    }
  }
}

With Claude Code:

claude mcp add nullify-acme-web -- nullify mcp serve --org acme-web

Scope findings to a specific repository with --repo:

{
  "mcpServers": {
    "nullify": {
      "command": "nullify",
      "args": ["mcp", "serve", "--repo", "my-repo"]
    }
  }
}

Configuration

The CLI stores configuration in ~/.nullify/:

Host resolution priority:

1. --host flag
2. NULLIFY_HOST environment variable
3. Config file (~/.nullify/config.json)

When the host is nullify.ai, the CLI signs in and refreshes tokens on nullify.ai and sends data requests to the selected organization's instance. GitHub Actions token exchange (--github-token) needs your instance host, not nullify.ai.

Global Flags

Requirements

• macOS, Linux, or Windows (amd64 or arm64)
• Docker — required only for local pentest scans (--local)
• A Nullify instance — request access

Documentation

Full documentation is available at docs.nullify.ai.

Contributing

Contributions are welcome. Please open an issue or submit a pull request.

License

MIT