RepoRescue MCP
Clones and inspects public GitHub repositories to extract evidence like manifests, dependencies, and version hints, and can run allow-listed repos in isolated Docker containers for reproducible verification.
README
RepoRescue MCP
RepoRescue turns a public GitHub URL into evidence that an agent can reason over. It can:
- clone a public GitHub repository with strict URL and size limits;
- identify Python manifests, declared dependencies, version hints, test runners and likely entry points;
- return source-linked evidence instead of invented environment facts;
- run allow-listed demonstration repositories inside constrained Docker containers;
- separate dependency installation from execution, then disable networking for the execution phase;
- produce bounded, redacted logs suitable for a verification report.
The service deliberately does not read a user's computer, accept arbitrary shell commands, execute private repositories, or execute repositories that are not explicitly allow-listed.
Local setup
py -3.11 -m venv .venv
.venv\Scripts\python -m pip install -e ".[dev]"
.venv\Scripts\pytest
docker build -f sandbox/Dockerfile.python311 -t repo-rescue-python:3.11 .
Run the Streamable HTTP MCP endpoint:
$env:REPO_RESCUE_ALLOWED_REPOS="pallets/click"
.venv\Scripts\repo-rescue-mcp
The MCP endpoint is exposed at http://localhost:8000/mcp by the official Python MCP SDK.
For a command-based managed MCP host, set REPO_RESCUE_TRANSPORT=stdio and run repo-rescue-mcp. The same executable defaults to Streamable HTTP for local or self-hosted use.
Tools
inspect_github_project
Read-only inspection for a public GitHub repository. Returns repository identity, commit SHA, bounded file tree, Python manifests, declared dependencies, version hints, risks and suggested verification commands.
reproduce_python_project
Runs an allow-listed Python repository in two Docker phases:
- dependency installation in an empty ephemeral workspace;
- execution in a separate container with networking disabled.
No host secrets or Docker socket are mounted into the target container. CPU, memory, process count, filesystem and runtime are constrained. The result contains actual exit codes and logs; success is never inferred.
The dedicated sandbox image contains only the baseline test runner and common terminal pager needed by the verified demo. Repository-specific dependencies are still resolved from the inspected manifest; the execution phase remains offline.
For a managed MCP host that cannot start Docker, set REPO_RESCUE_EXECUTION_BACKEND=direct. This fallback still accepts only explicitly allow-listed repositories and fixed verification commands, uses a temporary workspace, bounds time and output, and never exposes arbitrary shell execution. It is intentionally less isolated than the Docker backend and must not be enabled for untrusted public execution.
windows_environment_probe
Returns a copy-paste PowerShell command for users who need local-environment evidence. It only reads versions and package metadata.
Production boundary
Docker is adequate for an allow-listed competition demo, not for arbitrary hostile code. A public production service should replace Docker with a stronger isolation boundary such as gVisor or Firecracker, enforce outbound egress allow-lists during installation, scan uploaded archives, authenticate callers, rate-limit jobs and store no source after the job expires.
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。