GuardRail MCP v2.0 (Enterprise)
GuardRail MCP is an enterprise agent security server that provides multi-language code analysis, dependency scanning, and container security checks, with RBAC, policy packs, and audit logging.
README
GuardRail MCP v2.0 (Enterprise)
Multi-language enterprise agent security MCP: hybrid Python AST taint, multi-language sinks, parallel repo & PR scanning, OSV CVEs, SARIF/SBOM, Docker/K8s checks, plus multi-tenant RBAC, policy packs, audit logs, path sandbox, quotas, and Prometheus metrics.
Heuristic guardrails for developers & coding agents — not a replacement for commercial SAST or a formal SOC2 certification. Secrets are redacted in findings.
Enterprise guide: docs/ENTERPRISE.md
Earn on MCPize: docs/MCPIZE.md · mcpize.com/developers
List on more platforms: docs/DISTRIBUTION.md (Smithery, mcp.so, PulseMCP, Glama, official registry…)
Feature map
| Capability | Status | Entry point |
|---|---|---|
| Multi-language analysis | Shipped | audit_code_safety, languages.py |
| Repo-wide recursive scan | Shipped | scan-repo / scan_repository |
| Git diff / PR scanning | Shipped | scan-diff / scan_git_diff |
| Dependency & CVE (OSV) | Shipped | scan_dependencies |
| SARIF / SBOM | Shipped | export_sarif, generate_sbom |
| Fix drafts | Shipped | suggest_fixes |
| Docker/K8s security | Shipped | audit_container_config |
| Enterprise RBAC + API keys/JWT | Shipped | guardrail/enterprise/ |
| Policy packs (strict/soc2/pci) | Shipped | evaluate_policy, gateway gates |
| Path sandbox | Shipped | allowed_roots |
| Audit log + correlation IDs | Shipped | JSONL + list_audit_events |
| Rate limit + quotas | Shipped | per-tenant RPM / monthly |
| Prometheus metrics | Shipped | GET /metrics |
| K8s/Docker deploy | Shipped | deploy/ |
Language coverage
| Language | Engine |
|---|---|
| Python | Regex secrets + AST taint (eval aliases, SQL, shell, pickle, yaml) |
| JavaScript / TypeScript | eval/Function, child_process, innerHTML, TLS flags |
| Go | fmt.Sprintf SQL, shell Command, unsafe, InsecureSkipVerify |
| Java / Kotlin | Runtime.exec, concat SQL, ObjectInputStream, ScriptEngine |
| Rust | unsafe {}, shell Command, invalid certs |
| C / C++ | gets/strcpy/sprintf, system() |
| Docker / K8s / TF | Privileged, root, 0.0.0.0/0, latest tags, curl|sh, hostPath, … |
Install
cd guardrail-mcp
pip install -r requirements.txt
export PYTHONPATH=$PWD
CLI
# MCP server (IDE)
python -m guardrail --mode stdio
python -m guardrail --mode stdio-sdk # official mcp package
python -m guardrail serve --mode http --port 8787
# Full repo scan (parallel)
python -m guardrail scan-repo . --sarif out.sarif --fixes --deps --json-out report.json
# PR / git diff
python -m guardrail scan-diff . --base origin/main --head HEAD --sarif pr.sarif --fixes
# Dependencies + OSV (network)
python -m guardrail scan-deps . --osv
# SBOM
python -m guardrail sbom . --format both --out sbom.json
# SARIF only
python -m guardrail export-sarif . --out guardrail.sarif
Exit code 2 when security_verdict is REJECTED (handy for CI gates).
MCP tools (19)
Security: audit_code_safety · audit_cloud_cost · audit_infra_security · scan_repository · scan_git_diff · scan_dependencies · export_sarif · generate_sbom · suggest_fixes · audit_container_config · full_pipeline
Enterprise: enterprise_health · enterprise_policy_status · evaluate_policy · compliance_report · list_audit_events · issue_access_token · reload_enterprise_config · manage_tenant
Enable enterprise mode
export GUARDRAIL_ENTERPRISE=1
export GUARDRAIL_ENTERPRISE_CONFIG=$PWD/config/enterprise.json
python -m guardrail serve --mode http --port 8787
# Auth: Authorization: Bearer gr_demo_enterprise_key_change_me
GitHub Actions
Copy .github/workflows/guardrail.yml into your app repo (set PYTHONPATH / install path as needed). It:
- Scans PR diffs (or full repo on push)
- Writes
guardrail.sarif+ JSON report + SBOM - Uploads SARIF to GitHub code scanning (when permissions allow)
HTTP
GUARDRAIL_HTTP_API_KEY=dev python -m guardrail serve --mode http
curl -H "Authorization: Bearer dev" -H "Content-Type: application/json" \
-d '{"name":"scan_repository","arguments":{"path":"."}}' \
http://127.0.0.1:8787/v1/tools/call
Layout
guardrail/
safety.py / ast_engine.py / rules.py # Python hybrid core
languages.py # JS/Go/Java/Rust/C grids
repo_scan.py / git_scan.py # parallel repo + PR
deps.py / sbom.py # OSV + SBOM
sarif_export.py / fixes.py # SARIF + fix drafts
docker_k8s.py / infra_security.py # containers + IaC
pipeline.py / tools_catalog.py # orchestration
mcp_stdio.py / mcp_sdk_server.py / http_api.py
Tests
PYTHONPATH=. python -m unittest discover -s tests -v
Honest limits
- Non-Python “AST” is pattern-based, not tree-sitter/full compilers.
- OSV needs network; unpinned versions are inventoried but not queried.
- Fix drafts are template/agent prompts, not guaranteed patches.
- Not a substitute for CodeQL/Semgrep + SCA at enterprise scale — a strong local/CI agent guardrail.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。