pci-dss-mcp
pci-dss-mcp
MCP-сервер статического анализа для кодовых баз платежных сервисов на Go. Каждое обнаруженное нарушение PCI DSS v4.0.1 в кодовой базе платежного сервиса на Go сопоставляется с конкретным номером требования перед выпуском кода.
Что он делает
pci-dss-mcp — это stdio MCP-сервер, который запускает 12 сканеров, оркестратор и движок ИИ-триажа для кодовой базы платежного сервиса на Go. Каждая находка содержит requirement_id, сопоставленный с конкретным пунктом PCI DSS v4.0.1; см. docs/requirement-mapping.md для канонической таблицы правил и требований, а также testdata/vulnerable-payment-service/EXPECTED-FINDINGS.md для актуальных эталонных результатов.
Что pci-dss-mcp обнаруживает сегодня
Потоки ввода HTTP-фреймворков в приемники логов / ошибок / паник. Фреймворки уровня 1 (gin, chi, gorilla/mux, net/http (Go 1.22+), echo v4, fiber v2) и логгеры уровня 1 (log/slog, logrus, zap, zerolog, logr, klog, hclog) поддерживаются в v0.7. Уровень 2 (kratos, apex/log, charmbracelet/log) появится в v0.8. Уровень 3 (fasthttp, beego, iris, httprouter, внутренние проекты) настраивается пользователем через YAML в Phase 25 после выпуска. См. docs/http_input_taint.md.
Чем pci-dss-mcp НЕ является
Не замена широкому SAST. Используйте Semgrep, CodeQL или gosec для OWASP Top-10 и языково-независимых уязвимостей.
Не замена LLM-ревью кода. pci-dss-mcp сопоставляет специфичные для платежей проблемы с идентификаторами требований PCI DSS; LLM-агенты обнаруживают общие ошибки через рассуждения. Эти два уровня дополняют друг друга.
Не является языково-независимым. Именно специфичные для Go паттерны AST и отслеживание потоков загрязнения обеспечивают точность.
Не замена QSA. Статический анализ покрывает ~6% требований PCI DSS v4.0.1. Квалифицированный аудитор безопасности (QSA) должен подтвердить остальное.
Related MCP server: secureaudit-mcp
Установка
Установка Go (основной способ)
Требуется Go 1.25+:
go install github.com/shyshlakov/pci-dss-mcp@latestБинарный файл будет находиться по адресу $(go env GOPATH)/bin/pci-dss-mcp. См. docs/install-from-source.md для разрешения PATH, исправления codesign в macOS, проверки cosign и JSON-конфигурации MCP-клиента.
Docker (альтернатива)
docker pull ghcr.io/shyshlakov/pci-dss-mcp:v0.6.2Полезно для CI-пайплайнов, аудиторов QSA, которые не разрабатывают на Go локально, или любой среды без установленного инструментария Go.
Реестр MCP
Указан как io.github.shyshlakov/pci-dss-mcp на registry.modelcontextprotocol.io. Автоматически публикуется при каждом теге.
Использование
Добавьте в конфигурацию вашего MCP-клиента (Claude Desktop claude_desktop_config.json, Cursor .cursor/mcp.json или claude mcp add для Claude Code):
{
"mcpServers": {
"pci-dss-mcp": {
"command": "docker",
"args": ["run", "-i", "--rm",
"--mount", "type=bind,src=/Users/you/go/src,dst=/Users/you/go/src,readonly",
"ghcr.io/shyshlakov/pci-dss-mcp:v0.6.2"]
}
}
}src= и dst= отражают один и тот же абсолютный путь, чтобы контейнер видел ваш код по тому же пути, что и хост; промпты передают обычный путь хоста без трансляции. Для варианта go install и примеров для конкретных клиентов см. docs/usage.md.
Два промпта для вставки в ваш MCP-клиент:
Run pci-dss-mcp triage on /Users/you/payments-service. Use min_severity=MEDIUM and group findings by PCI DSS requirement.Generate a PCI DSS compliance report for /Users/you/payments-service in JSON format. Show requirement-level pass/fail status and severity counts.
Инструменты
Инструмент | Назначение | Документация |
| Все сканеры + ИИ-классификация + контекст файл:строка в одном вызове | |
| Отчет о прохождении/непрохождении требований (оркестратор всех сканеров) | |
| Хранение и логирование PAN/SAD (3.3.1, 3.4.1, 3.5.1) | |
| Слабое хеширование, захардкоженные ключи, обычный HTTP (4.2.1, 6.2.4) | |
| Небезопасные конфигурации TLS (4.2.1) | |
| Учетные данные в конфигурационных файлах (8.6.2) | |
| Ответы об ошибках, раскрывающие чувствительный контекст (6.2.4) | |
| Захардкоженные пароли, слабая политика, отсутствие MFA, подписи вебхуков (8.3.1, 8.3.6, 8.4.2, 8.6.2) | |
| Отсутствие аудиторских логов для платежных потоков (10.2.1) | |
| Отсутствие TTL, чувствительное хранение, отсутствие затирания (3.2.1, 3.3.1) | |
| Отсутствие CSP/SRI/nonce на платежных страницах (6.4.3, 11.6.1) | |
| Уязвимые зависимости Go через OSV (6.3.3); конфиденциальность в стиле govulncheck: имена модулей не отправляются в OSV.dev. См. docs/check_dependencies.md. Также включает | |
| CycloneDX 1.6 SBOM из go.mod/go.sum (6.3.2) | |
| Поиск требования PCI DSS v4.0.1 по ID |
Все инструменты объявляют типизированную OutputSchema. См. docs/tools.md для каталога и истории миграций.
Документация
docs/usage.md, настройка клиента, шаблоны промптов, подавление находок
docs/severity.md, модель серьезности и сопоставление правил с серьезностью
docs/taint.md, настройки анализа загрязнения по умолчанию и переключатели
docs/scoping.md, исключение пакетов и область CDE
docs/comparison.md, pci-dss-mcp против Semgrep / CodeQL / gosec / Snyk Code
docs/ci-cd.md, интеграция с GitHub Actions и GitLab CI
docs/pci-coverage.md, матрица покрытия требований PCI DSS v4.0.1
docs/install-from-source.md, сборка из исходников, проверка cosign, перезагрузка
docs/requirement-mapping.md, каноническая таблица rule_id к requirement_id
CONTRIBUTING.md, настройка разработки, цели фаззинга
ROADMAP.md, запланированные функции
CHANGELOG.md, история версий
Статус
Активная разработка, до v1.0. См. ROADMAP.md и CHANGELOG.md.
Лицензия
MIT, см. LICENSE.
pci-dss-mcp — это инструмент статического анализа. Он не может заменить квалифицированного аудитора безопасности (QSA). Используйте его результаты как входные данные для вашего процесса обеспечения соответствия, а не как само соответствие.
Available Tools
15 toolsaudit_log_coverageA
Scan Go source files for payment handlers missing structured audit logging (PCI DSS 10.2.1). Detects: missing logging, unstructured-only logging (fmt/log), and reports handlers with structured logging. Framework-aware: supports net/http, gin, echo handler signatures. Default: returns response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 3 per severity findings - plus a pagination.next_cursor for drill-down. Prefer this for mixed queries; min_severity / rule_filter drop to response_shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list. Use include_tests / exclude_patterns / min_severity / rule_filter for a filtered flat response.
Inputs: path (required), exclude_patterns, include_tests, include_untracked, +4 more. Returns: object
check_auth_strengthA
Scan Go source files for weak authentication: hardcoded passwords (PCI DSS 8.3.1), password policy with minimum length below 12 (PCI DSS 8.3.6), and payment routes missing MFA middleware (PCI DSS 8.4.2). Default: returns response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 3 per severity findings - plus a pagination.next_cursor for drill-down. Prefer this for mixed queries; min_severity / rule_filter drop to response_shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list. Use include_tests / exclude_patterns / min_severity / rule_filter for a filtered flat response. Maps findings to PCI DSS 8.3.1, 8.3.6, 8.4.2.
Inputs: path (required), exclude_patterns, include_tests, include_untracked, +4 more. Returns: object
check_data_retentionA
Scan Go source and config files for unsafe data retention: Redis/DB storage of CVV/PAN without TTL (PCI DSS 3.2.1), config files missing TTL on sensitive keys (PCI DSS 3.3.1), and incorrect memory zeroing timing after authorization. Scans .go, .yaml, .json, .toml files. Default: returns response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 3 per severity findings - plus a pagination.next_cursor for drill-down. Prefer this for mixed queries; min_severity / rule_filter drop to response_shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list. Use include_tests / exclude_patterns / min_severity / rule_filter for a filtered flat response. Maps findings to PCI DSS 3.2.1, 3.3.1.
Inputs: path (required), exclude_patterns, include_tests, include_untracked, +4 more. Returns: object
check_dependenciesA
Scan go.mod dependencies for known vulnerabilities (PCI DSS 6.3.3). Bulk-downloads the public OSV Go vulnerability snapshot and intersects locally against go.mod, matching the govulncheck privacy model. No module names are sent to OSV.dev. Cache TTL: 24h fresh, 24h-7d revalidate via ETag, >7d force-refresh. Run update_vulnerability_db first to bootstrap the cache for air-gapped environments. Default: returns response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 1 per severity findings - plus a pagination.next_cursor for drill-down. Prefer this for mixed queries; min_severity / rule_filter drop to response_shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list. Use min_severity / rule_filter / positive limit for a filtered flat response. Maps findings to PCI DSS 6.3.3.
Inputs: path (required), mode, cursor, limit, +2 more. Returns: object
check_encryptionA
Scan Go source files for encryption violations: weak hash algorithms (md5/sha1) with context scoring, hardcoded encryption keys/IVs, and plain HTTP URLs. Default: returns response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 3 per severity findings - plus a pagination.next_cursor for drill-down. Prefer this for mixed queries; min_severity / rule_filter drop to response_shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list. Use include_tests / exclude_patterns / min_severity / rule_filter for a filtered flat response. Maps findings to PCI DSS 6.2.4, 4.2.1.
Inputs: path (required), exclude_patterns, include_tests, include_untracked, +4 more. Returns: object
check_error_handlingA
Scan Go source files for payment handler error disclosure: detects err.Error() in http.Error, fmt.Fprintf with error variables, w.Write with error bytes, and json.Encode of errors. Default: returns response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 3 per severity findings - plus a pagination.next_cursor for drill-down. Prefer this for mixed queries; min_severity / rule_filter drop to response_shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list. Use include_tests / exclude_patterns / min_severity / rule_filter for a filtered flat response. Maps findings to PCI DSS 6.2.4.
Inputs: path (required), exclude_patterns, include_tests, include_untracked, +4 more. Returns: object
check_payment_page_scriptsA
Scan Go source files and HTML templates for payment page script security violations (PCI DSS 6.4.3, 11.6.1). Detects: missing Content-Security-Policy headers in Go payment handlers, unsafe-inline/unsafe-eval in CSP, external scripts without SRI (integrity attribute) in HTML templates, inline scripts without nonce attribute. Framework-aware: supports net/http, gin, echo handler signatures. Default: returns response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 3 per severity findings - plus a pagination.next_cursor for drill-down. Prefer this for mixed queries; min_severity / rule_filter drop to response_shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list. Use include_tests / exclude_patterns / min_severity / rule_filter for a filtered flat response. Maps findings to PCI DSS 6.4.3, 11.6.1.
Inputs: path (required), exclude_patterns, include_tests, include_untracked, +4 more. Returns: object
check_secrets_in_configsA
Scan configuration files (.env, .yaml, .json, .toml) for hardcoded secrets: API keys, passwords, tokens, and connection strings with embedded credentials. Default: returns response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 3 per severity findings - plus a pagination.next_cursor for drill-down. Prefer this for mixed queries; min_severity / rule_filter drop to response_shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list. Use include_tests / exclude_patterns / min_severity / rule_filter for a filtered flat response. Maps findings to PCI DSS 8.6.2.
Inputs: path (required), exclude_patterns, include_tests, include_untracked, +4 more. Returns: object
check_tls_configA
Scan Go source files for TLS configuration violations: InsecureSkipVerify, weak MinVersion (below TLS 1.2), missing MinVersion, and prohibited cipher suites (RC4, 3DES, NULL). Default: returns response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 3 per severity findings - plus a pagination.next_cursor for drill-down. Prefer this for mixed queries; min_severity / rule_filter drop to response_shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list. Use include_tests / exclude_patterns / min_severity / rule_filter for a filtered flat response. Maps findings to PCI DSS 4.2.1.
Inputs: path (required), exclude_patterns, include_tests, include_untracked, +4 more. Returns: object
explain_requirementA
Look up a PCI DSS v4.0.1 requirement by ID. Returns title, description, and testing procedure.
Inputs: requirement_id (required). Returns: requirement
generate_compliance_reportA
Raw PCI DSS v4.0.1 compliance report without AI triage — intended for CI gates, audit artifacts, and requirement-level pass/fail lists. For interactive "scan this project" prompts call triage_findings instead. Default unfiltered call returns a compact summary (metadata, totals, requirement_statuses, top 20 findings per severity, cursor for follow-up). Supply min_severity / rule_filter / limit to get a paged flat list (60 per page with cursor), or cursor= to resume a prior session (10-minute TTL). min_severity / rule_filter drop the response to shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Taint analysis is ON by default; set include_taint=false for fast dev iteration.
Inputs: path (required), dep_scan_mode, include_tests, include_taint, +4 more. Returns: object
generate_sbomA
Generate a CycloneDX v1.6 SBOM for a Go project. Default behavior: writes sbom.json (or sbom.xml when format=xml) next to the scanned go.mod and returns metadata only (output_path, size_bytes, component_count, unknown_licenses). Override the destination with output_path (must be absolute). Pass inline=true to return the serialized SBOM in the MCP response instead (capped at 64 KB; returns SBOM_TOO_LARGE above that). Parses go.mod + go.sum offline against the local GOMODCACHE; cache-miss modules surface as UNKNOWN-LICENSE. Satisfies PCI DSS 6.3.2 (software inventory, mandatory since March 2025).
Inputs: path (required), format, output_path, inline, +2 more. Returns: mode, bom_format, spec_version, component_count, +9 more
scan_pan_dataA
Default: returns response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 3 per severity findings - plus a pagination.next_cursor for drill-down. Prefer this for mixed queries; min_severity / rule_filter drop to response_shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list. Use include_tests / exclude_patterns for a filtered flat response. Maps findings to PCI DSS 3.3.1, 3.4.1, 3.5.1.
Inputs: path (required), exclude_patterns, include_tests, include_untracked, +5 more. Returns: object
triage_findingsA
Run all PCI DSS v4.0.1 scanners + AI-assisted prioritization + file:line enrichment on a Go project in a single call. Default: response_shape "summary" with by_severity counts, a capped by_rule histogram (top 10 + more_rules), and top 1 per severity enriched finding, plus pagination.next_cursor for drill-down. min_severity / rule_filter drop the response to shape "flat" but still carry summary.by_severity + summary.by_rule for full-scan context. Follow the cursor for the full paginated list.
Inputs: path (required), dep_scan_mode, include_tests, include_taint, +4 more. Returns: object
update_vulnerability_dbA
Download fresh OSV Go vulnerability snapshot to local cache for offline scanning. Downloads from gs://osv-vulnerabilities/Go/all.zip (~7.5MB). This is the ONLY tool that makes network requests. Cache stored at PCI_MCP_CACHE_DIR or ~/.pci-dss-mcp/vuln-cache/ by default.
Inputs: output_path. Returns: cache_path, vuln_count, download_size_bytes, previous_cache_date, +1 more
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 15 tools
Each tool targets a distinct security check or utility: audit log coverage, authentication strength, data retention, dependencies, encryption, error handling, payment page scripts, secrets, TLS, requirement lookup, compliance report, SBOM, PAN data scan, triage, and vulnerability DB update. No two tools have overlapping purposes.
Tool names follow a clear verb_noun pattern (e.g., check_*, generate_*, explain_*, update_*, scan_*, triage_). All use consistent snake_case, and the verb clearly indicates the action. The slight deviation of 'audit_log_coverage' is minor and still descriptive.
15 tools is well-scoped for a PCI DSS compliance server. It covers a wide range of checks (authentication, encryption, secrets, etc.) plus utilities (SBOM, report, requirement lookup), without being excessive or too few.
The tool set comprehensively covers PCI DSS requirements relevant to Go source code and configuration, including authentication (8.3), data retention (3.2, 3.3), encryption (4.2, 6.2), error handling (6.2), TLS (4.2), secrets (8.6), dependencies (6.3), and software inventory (6.3). The triage and report tools tie everything together.
Maintenance
Related MCP Connectors
Zero-config MCP security scanner for AI-generated apps. 25K+ vulnerability patterns.
Security scanner for MCP servers. Detect vulnerabilities, prompt injection, and tool poisoning.
IaC, CI and dependency security review for coding agents: local scan CLI plus hosted OAuth MCP.
Static MCP server scan: injection, secrets, tool poisoning. SARIF + rules catalog.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceGo-based SBOM, vulnerability, and secret scanner with MCP support.3Apache 2.0
- AlicenseAqualityCmaintenanceAn AppSec-focused MCP server that performs static analysis scans on C/C++ source code for memory-safety vulnerabilities and parses compiled PE/ELF binary headers locally to audit active defensive compiler mitigations (ASLR, DEP/NX, PIE).41MIT
- AlicenseAqualityDmaintenanceAgent-native "safe to ship?" security gate for AI-generated code. Uses real parsers and inter-rocedural taint analysis (JS/TS, Python, Go) to flag the classes AI coding agents get wrong — secrets, SQL injection, SS, SSRF, path traversal, command injection, weak JWT/CORS — and ranks findings by confidence. Exposes a scan tool over MCP.18 npm2MIT
- AlicenseAqualityDmaintenanceA production-grade security auditing MCP server that wraps semgrep (SAST) and gitleaks (secret detection) to enable one-click code security scanning via MCP stdio protocol.1MIT