SAST · Secrets · SCA

Real findings, not noise.
Built for AI agents.

A free, Semgrep-compatible security scanner. It combines tree-sitter AST analysis and cross-file taint tracking, 270+ secret detectors, and software composition analysis to find real vulns, leaked secrets, and vulnerable dependencies in a single pass. A verified autofix harness then closes them out, all from one binary that drops into CI or runs inside your coding agent over MCP.

Get Started

Bash

$ npx @mondoohq/xgrep scan ./src

  287 files · 1,000+ rules · secrets on

  CRITICAL  sql-injection    src/db/query.py:42   request.args → cursor.execute
  HIGH      aws-access-key   src/config.py:11
  WARNING   weak-crypto      src/auth/hash.py:28

  3 findings · run `xgrep fix` to remediate, or `xgrep deps why` to trace dependencies

Three scanners. One tool.

xgrep unifies static analysis, secret detection, and software composition analysis — no separate tools, no runtime, no warm-up.

SAST→

Tree-sitter AST matching with cross-file taint analysis. Semgrep-compatible, accuracy-first.

Secrets→

270+ detectors across 150+ providers, with deep git-history scanning and live validation.

SCA→

Dependency inventory, reachability, and CVE matching in the same pass, with CycloneDX/SPDX SBOM and CBOM included.

SAST that reports what's actually exploitable.

xgrep parses your code with tree-sitter and traces untrusted input from source to sink across files and functions. When it flags a vulnerability, it's real — not a style nit.

AST, not regex

Tree-sitter parsers understand structure and typed metavariables — far fewer false positives than text matching.

Cross-file taint analysis

Follows tainted data through calls and modules to confirm a real source-to-sink path before reporting.

vuln vs. audit tiers

Exploitable issues are separated from hardening advice, so severity reflects real risk — not noise.

Vulnerability coverage

✓Injection (SQL, command, code, LDAP, template)

✓Cross-site scripting (XSS)

✓Server-side request forgery (SSRF)

✓Insecure deserialization

✓Path traversal

✓Authentication & session flaws

✓Weak crypto & TLS

✓ReDoS & denial of service

✓Sensitive-data exposure

Every rule is mapped to CWE and the OWASP Top 10, and every finding ships with code context and remediation guidance. Infrastructure as code is covered too — Terraform/HCL, Dockerfiles, and GitHub Actions workflows.

Bring your own rules — Semgrep-compatible YAML works unchanged

YAML

rules:
  - id: sql-injection
    patterns:
      - pattern: cursor.execute($QUERY)
      - pattern-not: cursor.execute("...")
    message: Possible SQL injection via string formatting
    severity: CRITICAL
    languages: [python]

33 languages supported

BashCC++C#ClojureDartDockerfileElixirErlangGoHCLHTMLJavaJavaScriptJSONJuliaKotlinLispLuaOCamlPHPPythonRRubyRustScalaSchemeSoliditySwiftTSXTypeScriptXMLYAML

Catch secrets before they ship — and the ones already leaked.

Secret scanning runs on every scan. 270+ provider-specific detectors match real key formats — prefixes, lengths, checksums — instead of guessing on entropy, so you get signal, not noise.

270+ detectors, 150+ providers

AWS, GitHub, Stripe, OpenAI, Anthropic, Slack, Cloudflare, and hundreds more.

Scans full git history

Find secrets that were committed and later removed, with the commit and author that introduced them.

Validates live credentials

Optionally calls the provider to mark each secret confirmed_live, revoked, or invalid.

Secrets hide in git history and encoded blobs — xgrep scans both

Bash

# Find secrets across full git history, with commit provenance
xgrep scan --history --category secrets .

# Decode base64/hex/gzip payloads to catch hidden keys
xgrep scan --decode --category secrets .

# Confirm which leaked keys are still live
xgrep scan --validate --category secrets .

Well-known placeholder values (like AWS’s documentation keys) are filtered out, so examples never page you. Secrets are reported even in test paths, where real credentials most often slip through.

AWSAzureGoogle CloudGitHubGitLabStripeOpenAIAnthropicSlackDiscordTwilioSendGridDatadogCloudflareVaultMongoDBSnykVercelJWTPrivate keys+150 more

Software composition analysis, in the same pass.

xgrep inventories your dependencies straight from lockfiles, traces which of them your code actually reaches, and flags the vulnerable ones. A single xgrep scan covers your first-party code and your dependencies at once.

What’s in here?

A standards-compliant SBOM across 11 ecosystems (Go, npm, Python, Rust, Ruby, PHP, Java, .NET, Dart, Elixir, Swift), emitted as CycloneDX or SPDX with PURLs, CPEs, and licenses. Fully offline: nothing is installed or executed.

What actually uses it?

deps why traces every package back to the first-party files that import it, over the same code graph your agents query for reachability. A dependency nobody imports is a very different risk.

What’s vulnerable?

The same run matches dependencies against known CVEs, so vulnerable packages surface alongside code findings. (CVE matching calls the Mondoo vulnerability service; SBOM and deps queries stay fully offline.)

One toolchain for inventory, reachability, and vulnerabilities

Bash

# A standards-compliant SBOM, straight from lockfiles
xgrep sbom --format cyclonedx-json -o sbom.json .

# Which of our own code actually reaches this package?
xgrep deps why --package requests .

# SAST plus dependency CVEs, in a single pass
xgrep scan .

GonpmPythonRustRubyPHPJava.NETDartElixirSwift

Cryptography Bill of Materials (CBOM)

Inventory the cryptography in your code for post-quantum migration. Run xgrep sbom --cbom . for a CycloneDX 1.6 CBOM.

Your CI pipeline shouldn't wait for your scanner.

No interpreter to boot, no packages to install, no warm-up, and no AI agent or hosted backend required. SAST, secrets, and SCA all run from one binary the moment CI calls it.

60 ms cold start

semgrep needs over a second before scanning even starts

Single binary

Zero dependencies. No Python runtime, no Docker.

Parallel by default

Scales to all CPU cores

Findings your agent can actually fix, verified.

Discovery is only half the job. xgrep hands your agent a fix contract for each finding and verifies the result, so the loop closes with a real remediation instead of a suggestion. scan discovers and never mutates source; fix is the only command that writes.

Deterministic

One provably-correct edit (for example verify=False → verify=True). Applied automatically, no agent required.

Assisted

xgrep emits a fix contract describing exactly what a valid fix must satisfy, and your coding agent writes the edit to meet it.

Advisory

Where no safe automated edit exists, xgrep returns structured, natural-language remediation guidance instead.

Every edit is verified before it lands

A fix passes a parse-clean gate, an atomic write, and a re-scan gate that rejects it unless it actually cleared the finding without introducing an equal or higher severity issue. xgrep never writes an unverified fix.

Agents drive the same loop over MCP as fix_verify and fix_apply. The interactive TUI (bare xgrep fix) walks findings, delegates assisted fixes to your coding agent, pulls graph context with g, and opens a pull request for applied fixes with P.

Preview, apply, or review interactively. The fix always re-scans clean.

Bash

# Preview diffs without touching disk
xgrep scan --json ./src | xgrep fix verify

# Write the accepted edits
xgrep scan --json ./src | xgrep fix apply

# Or review every finding in the interactive TUI
xgrep fix

Built for the AI-native security workflow.

Traditional SAST dumps 500 findings. Your team ignores 400 of them. xgrep lets AI agents triage with full code context.

Traditional SAST

1Scan codebase

2Dump 500 findings

3Human triages for days

4Next sprint, repeat

AI-Native SAST

1xgrep scans the codebase

2Agent skills trace call chains via the code graph

3Auto-classifies TP / FP with evidence

4Generates fixes with full context

xgrep-triage

Investigate and classify scan findings using code-graph call-chain and dataflow analysis

xgrep-inspect

Navigate and understand source code with AST-powered code intelligence

xgrep-rule-creator

Create custom detection rules with a test-first methodology

xgrep-remediate

Safely fix confirmed findings with the verify/apply autofix harness

secure-coding

Proactively avoid generating vulnerable code patterns while writing

Bash

# Install the bundled skills into Claude Code
npx @mondoohq/xgrep skill install

“xgrep does not embed an LLM. It gives your agent skills and tools.”

Works with your AI agent.

xgrep integrates as a backend skill for any coding agent — via MCP, CLI, or direct tool calls. Your agent scans, traces, and fixes without leaving the session.

Claude Code

Cursor

OpenAI Codex

Gemini CLI

GitHub Copilot

Windsurf

Goose

Cline

Kiro

Roo

Trae

OpenCode

Kilo Code

Claude Code

Cursor

OpenAI Codex

Gemini CLI

GitHub Copilot

Windsurf

Goose

Cline

Kiro

Roo

Trae

OpenCode

Kilo Code

First scan in 30 seconds.

Add xgrep as a backend skill for your coding agent, or run it directly.

Install as agent skill

Bash

# Claude Code — install the bundled skills
# (triage, inspect, rule-creator, remediate, secure-coding)
npx @mondoohq/xgrep skill install

# Cursor / Windsurf / any MCP client
{
  "mcpServers": {
    "xgrep": {
      "command": "npx",
      "args": ["@mondoohq/xgrep", "mcp"]
    }
  }
}

Run directly

Bash

# npx (zero install)
npx @mondoohq/xgrep scan ./src

# Or install globally
npm install -g @mondoohq/xgrep

# Rules are built in — just point at your code
xgrep scan ./src

Common commands

Bash

# Scan for vulnerabilities and secrets
xgrep scan .

# Diff-aware scan in CI (auto-detects the provider)
xgrep ci

# Remediate findings with the verified autofix harness
xgrep fix

# Generate a CycloneDX SBOM, or a Cryptography BOM
xgrep sbom --format cyclonedx-json -o sbom.json .
xgrep sbom --cbom .

# Trace which of your code reaches a dependency
xgrep deps why --package requests .

# Scan git history for leaked secrets
xgrep scan --history --category secrets .

# Output SARIF for GitHub Code Scanning
xgrep scan --sarif -o results.sarif .

In your editor

The Mondoo VS Code extension runs xgrep’s language server for real-time, in-editor diagnostics, taint visualization, and quick fixes. Available on both the Visual Studio Marketplace and Open VSX (for VSCodium, Cursor, and Windsurf).

VS Code Marketplace Open VSX

Real findings, not noise.Built for AI agents.

Three scanners. One tool.

SAST that reports what's actually exploitable.

Vulnerability coverage

33 languages supported

Catch secrets before they ship — and the ones already leaked.

Software composition analysis, in the same pass.

Your CI pipeline shouldn't wait for your scanner.

Findings your agent can actually fix, verified.

Built for the AI-native security workflow.

Traditional SAST

AI-Native SAST

Works with your AI agent.

First scan in 30 seconds.

Install as agent skill

Run directly

Common commands

In your editor

Real findings, not noise.
Built for AI agents.