Codex Security Cloud MCP
MCP server for the Codex Security Cloud service, enabling agents to list, inspect, close, reopen findings, request PRs, and apply generated patches.
README
Codex Security Cloud MCP
MCP server for the Codex Security Cloud service exposed by ChatGPT/Codex Cloud.
The server assumes the user already has Codex installed and authenticated. It reads the existing Codex auth file and does not implement a separate login flow.
Tools
list_security_metadata: lists accessible repos and author filters.list_findings: lists open or closed findings with repo, severity, patch, author, sort, cursor, and limit filters.get_finding: fetches normalized finding detail, including generated patch metadata when present.close_finding: closes a finding asfixed,wontfix,duplicate, orfalse_positive.reopen_finding: reopens a closed finding asnew.request_site_pr: asks the cloud service to start its site-side patch/PR workflow, then briefly polls the finding for PR URL/state metadata.apply_generated_patch: applies an already-generated patch locally, commits it, and optionally closes the finding as fixed.
Use the findingId returned by list_findings for get_finding, close/reopen, PR, and patch tools. The MCP intentionally omits backend-only ids and raw payloads from normal responses.
list_findings is intentionally compact so an agent can choose what to inspect without loading every finding body. get_finding is also compact by default: it includes a derived summary, compact relevant-line locations, generated patch metadata, and site PR metadata. The summary is extracted from an explicit Summary/Overview section when present, otherwise from the first paragraph with a bounded length. Use includeDescription: true when an agent needs the exact full description from the endpoint. Use includePatchDiff: true only when an agent needs to inspect or manually repair the generated patch. Use includeEvidence: true only when an agent needs relevant-line source content, validation/fix reports, or attack-path evidence. apply_generated_patch fetches the generated diff internally, so applying a clean generated patch does not require loading the diff into model context.
apply_generated_patch fails before applying if the generated patch is missing, the repository has staged changes, the patch touches dirty files, or git apply --check fails. On failure it returns the phase, a compact finding reference, current HEAD, expected base commit, touched files, git stderr, and the suggested commit message so an agent can take over manually.
If autoClose is true, the close happens only after a successful local commit. The close reason uses neutral wording:
An agent applied and committed the generated patch on <date time>.
Commit: <sha>
Development
npm install
npm run build
node ./dist/server.js
MCP Configuration
{
"mcpServers": {
"codex-security-cloud": {
"command": "node",
"args": ["/Users/obinnanwachukwu/Code/codex-security-cloud-mcp/dist/server.js"]
}
}
}
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。