leveret
Self-hosted MCP engine for private code reviews, providing deterministic static analysis and AST-level search over diffs, with findings passed to a review agent of your choice.
README
leveret
A leveret is a young hare — small, fast, and born with its eyes open.
leveret is a self-hosted, hybrid engine for private code reviews: the successor to hosted AI review bots for teams whose code stays home. Deterministic static analysis, AST-level search, and a graded filtering pipeline exposed over MCP, driven by whatever reviewing agent you bring (BYOM — Claude Code, a local model via any MCP-capable client, or a CI runner). The engine layer itself never calls an LLM.
Pipeline the project targets:
[diff] → [deterministic first pass: leveret scan] → [review agent: leads → concerns]
→ [verification agent: refute-or-evidence filter] → [report]
Tools
- scan
{repo, base? | files?, engines?, delta?}— runs the engines applicable to the change set and returns normalized findings (engine, rule, severity, file, line, message, provenance) plus a per-engine status report. With a base ref the scan is delta by default: findings already present at the base tree are dropped as pre-existing (counted, never silent) and survivors are taggedintroduced. Engines: semgrep (registry security + per-language rulesets), gitleaks (secrets overbase..HEADcommits), shellcheck, ruff, actionlint, zizmor (workflow security), osv-scanner (lockfile CVEs), typos (spelling), jscpd (profile-gated duplication), profile-declared semgrep/ast-grep rule packs, and arbitrary SARIF-emitting commands viacustom:profile entries (recipes: psalm taint, hadolint, trivy, ...). Findings are review leads, not verdicts. - ast_search
{repo, pattern, lang, paths?}— structural pattern matching via ast-grep (metavariables, syntax-aware), for call-site-shaped questions text grep gets wrong. - context
{repo, files}— prioritization signal, not findings: per-function cyclomatic complexity (lizard), 12-month churn, last-touched date.
Run
npm install && npm run build
node dist/server.js # stdio MCP server
npm test # integration tests (need semgrep, gitleaks, shellcheck, ruff, actionlint, ast-grep on PATH)
Claude Code registration:
claude mcp add leveret -- node /Users/andre/git/leveret/dist/server.js
Status / roadmap
See DESIGN.md — architecture, the three-grade filter (actionable / priced-noise / false-positive), the in-repo memory store, agent prompt contracts, and the validation benchmark gating adoption.
License
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。
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。