OmniCodex
Windows-local MCP gateway for the Codex runtime, enabling owner-authenticated tool access via Streamable HTTP with optional Oracle consent companion.
README
OmniCodex
OmniCodex is a Windows-local, owner-authenticated Streamable HTTP MCP gateway for the callable tool surface of an already installed Codex runtime. It owns a separate hidden child runtime and never registers itself in the Codex desktop app, modifies Codex authentication/configuration, or controls the desktop app.
Install and configure
Requirements: Windows, Node.js 22.12 or newer, and an installed Codex runtime.
npm install --global @heelee912/omnicodex
omnicodex init --issuer https://tenant.example/ --audience https://mcp.example `
--resource https://mcp.example --subject owner-sub
omnicodex auth login # official Auth0 device flow; refresh token stays in Windows keyring
omnicodex auth setup # idempotently creates the dedicated API and strict DCR grant
omnicodex start
omnicodex status
omnicodex runtime probe --json # isolated hidden runtime + full-surface live check
Initialization is fail-closed. OAuth issuer, audience/resource, scopes, and the
owner sub must match. auth login downloads a pinned, checksum-verified
official Auth0 CLI when one is not supplied and uses its Windows-keyring-backed
device login. No Management API token is written to OmniCodex configuration.
OmniCodex binds locally to 127.0.0.1; remote access
requires an explicitly configured authenticated HTTPS ingress.
Optional Oracle consent companion
The optional companion attaches only to an explicitly configured loopback
Chrome DevTools endpoint. It never launches, profiles, navigates, or terminates
Chrome. Setup is a dry-run unless --execute is supplied.
omnicodex oracle setup --connector-id exact-id --connector-name OmniCodex `
--run-id exact-run --resource https://mcp.example/mcp --surface /mcp/full `
--cdp-endpoint http://127.0.0.1:9222 --non-interactive
omnicodex oracle status --json
omnicodex oracle test # no click
omnicodex oracle test --execute # Connect / Always allow only
omnicodex oracle disable --execute
It is bound to the exact app, run, session, resource, and surface. General, one-time, destructive, stale, ambiguous, or reconnected confirmations fail closed. Receipts contain identifiers and state hashes only—never prompts, screenshots, secrets, or raw tool content.
Architecture and safety
- The installed runtime is the authority for callable tools.
- Ordinary calls do not silently become model calls.
- Protected Codex files are read-only and hash-checked by the non-interference guard.
- Owned console children use hidden/no-window creation.
- OAuth resource, issuer, scope, and owner subject are checked before MCP sessions.
- Browser automation is optional and separate from MCP server authority.
Operational commands include doctor, logs, runtime discover, runtime probe, autostart,
auth, tunnel, update, and rollback. Run omnicodex --help for details.
Documentation: 한국어 · 日本語 · 中文 · source attribution.
Release policy
Maintainers run npm run check, npm run build, and npm pack --dry-run.
Publishing is intentionally separate: npm publish --provenance --access public
must run only from the protected GitHub release workflow after review. No local
release script publishes, pushes, deploys, or changes GitHub state.
MIT licensed. This community project is not an official OpenAI product.
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。