D'AUBE // BRIDGE²

io.github.daubesonntag-dotcom由 GitHub 帳號 daubesonntag-dotcom 發布v0.1.3更新於 Oct 24, 2026

Universal AI skill compiler and MCP bridge. Write once, run across agents.

已驗證STDIO僅桌面開發者工具AI 與機器學習

概覽

AI 產生的概覽

把一份可攜式技能規格編譯成面向多個 AI 代理與 MCP 主機的整合產物，並以 MCP 工具形式提供驗證與編譯能力。

可以做什麼
D'AUBE // BRIDGE² 是一個通用 AI 技能編譯器兼 MCP 橋接器。它從單一可攜式技能定義產生面向 MCP 主機、OpenAI/ChatGPT/Codex、Claude、Gemini、DeepSeek、Meta/Llama 以及 Chromium 瀏覽器的整合產物。其內建 MCP 伺服器提供三個基礎工具：bridge_targets 用來探索支援的編譯目標，validate_skill 用來驗證可攜式技能物件，compile_skill 用來在記憶體中為其他代理或工具編譯技能。
適用情境
當同一項能力需要發佈到多個代理平台，而你不希望把它反覆改寫成 MCP 工具、各廠商函式結構、技能、外掛與瀏覽器設定時，適合使用它。它適合維護跨廠商工具鏈、並希望產物保持可讀、可比對的團隊。
執行需求
以本機程序在使用者機器上執行，僅限桌面端，使用 stdio 傳輸。README 透過 Python 與 pip 安裝（包含來自 GitHub Releases 的 wheel），並提到使用 uv 執行環境的 MCPB 打包；也可在本機主機上提供 HTTP MCP 端點。未宣告任何驗證、環境變數或標頭。
安裝前請注意
README 的外部審查說明指出該部署紀錄通過 AI Deploy Network 審查，並不代表經客戶確認的商業成果，也不保證正式環境的表現。安裝會從 GitHub 儲存庫與發佈 wheel 取得程式碼，HTTP 模式會暴露一個本機端點。README 表示驗證與敏感操作維持明確，但清單中未宣告任何憑證。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 D'AUBE // BRIDGE²，將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Desktop only，透過 STDIO。 STDIO 服務會啟動本機處理程序，因此需要 SourceWeft 桌面主機。

其他 MCP 客戶端

參照 儲存庫 中的啟動說明。

README

D'AUBE // BRIDGE²

Write once. Run across agents.

D'AUBE // BRIDGE² is an open-source universal AI skill compiler and MCP bridge. Define one portable capability spec, then compile it into integration artifacts for ChatGPT/Codex, Claude, Gemini, DeepSeek, Meta/Llama, MCP hosts, and Chromium browsers.

[Release] [MCP Registry] [License: MIT] [Python]

External review

On 2026-09-18, the BRIDGE² deployment record was AI Deploy Network Approved after review of the submitted implementation evidence. This means the deployment record passed AI Deploy Network review; it is not a client-confirmed commercial outcome or a guarantee of production performance.

The canonical publication narrative is maintained in docs/AI_DEPLOY_CASE_STUDY.md. The reviewed case study is now publicly available on AI Deploy Network.

Why BRIDGE²

AI tooling is fragmented. The same capability is repeatedly rewritten as MCP tools, provider function schemas, skills, plugins, and browser configuration. BRIDGE² treats those surfaces as compiler targets while keeping capability logic provider-neutral.

text
skill.yaml  |  +-- MCP tool schema  +-- OpenAI / ChatGPT / Codex artifacts  +-- Claude plugin + SKILL.md  +-- Gemini SKILL.md + MCP config  +-- DeepSeek tool schemas  +-- Meta/Llama function schemas  +-- Chromium extension manifest

Current release: 7 target families, 15 deterministic artifacts, one MCP server.

30-second start

bash
git clone https://github.com/daubesonntag-dotcom/daube-agent-bridge.gitcd daube-agent-bridgepython -m pip install -e ".[dev]"daube-bridge validate examples/skill.yamldaube-bridge compile examples/skill.yaml -o build/sample

Install the verified v0.1.3 wheel directly from GitHub Releases:

bash
python -m pip install https://github.com/daubesonntag-dotcom/daube-agent-bridge/releases/download/v0.1.3/daube_agent_bridge-0.1.3-py3-none-any.whlpython -m daube_bridge doctor

python -m daube_bridge ... is the shim-free CLI path for Windows environments where Application Control blocks generated console-script executables.

Run BRIDGE² as a local stdio MCP server:

bash
python -m daube_bridge.server

Or expose the HTTP MCP endpoint for remote/local integrations:

bash
daube-bridge serve# http://127.0.0.1:8000/mcp

One-click MCPB

BRIDGE² ships an MCPB v0.4 bundle definition using the cross-platform uv runtime. Build the deterministic bundle with:

bash
python scripts/build_mcpb.py

The build is reproducible: identical source produces an identical archive hash. Official MCP Registry metadata lives in server.json.

Target matrix

SurfaceOutputStrategy
MCP hostsMCP server + tool schemasShared transport/tool backbone
OpenAI / ChatGPT / CodexSKILL.md, MCP descriptor, function schemasSkills + MCP/tool calling
Claude.claude-plugin/plugin.json, .mcp.json, SKILL.mdPlugin + portable skill + MCP
Gemini / Google agentsSKILL.md, MCP configPortable skill + MCP
DeepSeekOpenAI-style tools.jsonFunction/tool calling adapter
Meta / Llamaneutral tools.jsonApplication/router function schema
Chromium browsersManifest V3 side panelEndpoint/config control surface

This matrix describes integration artifacts BRIDGE² generates. It does not invent provider capabilities that do not exist.

Portable skill format

yaml
name: My Research Skillversion: 0.1.0description: Research and compare options.tools:  - name: compare_options    description: Compare options against explicit criteria.    parameters:      type: object      properties:        options:          type: array          items: {type: string}      required: [options]

MCP tools

The bundled server exposes three small primitives:

  • bridge_targets — discover supported compilation targets.
  • validate_skill — validate a portable skill object.
  • compile_skill — compile a skill in-memory for another agent or tool.

That makes BRIDGE² usable both as a CLI and as infrastructure callable by other agents.

Verify before release

bash
python scripts/verify.pypython scripts/build_mcpb.py.tools/mcp-publisher.exe validate

The verification gate covers tests, Ruff, target discovery, sample compilation, wheel creation, deterministic MCPB creation, and official MCP Registry metadata validation.

After publishing a GitHub Release, smoke-test the public wheel in an isolated environment:

bash
python scripts/smoke_release.py 0.1.3

Roadmap

  • ChatGPT Apps SDK interactive widget target.
  • npx / TypeScript compiler parity.
  • Provider capability detection and compatibility linting.
  • Signed bundles and provenance manifests.
  • Remote registry + searchable skill catalog.
  • One-click adapters for agent IDEs and orchestration frameworks.
  • Golden interoperability fixtures across providers.

Project principles

  1. Portable by default. Prefer MCP and open schemas over provider lock-in.
  2. Evidence over pretending. Never claim native support that a provider does not have.
  3. Readable outputs. Generated artifacts stay human-auditable and diffable.
  4. Safe boundaries. Authentication and sensitive actions remain explicit.
  5. Fast adoption. A useful first result should take minutes, not a framework migration.

Contributing

Issues, adapters, compatibility fixtures, docs, and provider integrations are welcome. Provider-format changes should include a link to authoritative documentation and a minimal reproducible fixture.

Useful starting points:

License

MIT © 2026 D'AUBE SONNTAG.

來源：README.md，提交 56af43f

工具

0
SourceWeft 尚未在沙箱中啟動過該軟體包，因此其工具未知。

版本歷史

1
  1. v0.1.3最新Sep 16, 2026

評分

已滿足適用檢查項的 50%

該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通25 分

    請保持端點能夠回應 MCP 握手；SourceWeft 每 30 天探測一次。

  • 已宣告認證方式10 分

    請在註冊中心條目中宣告憑證請求標頭，或保持端點能夠回應探測，以便判定其認證方式。

  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list，或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分