cbb
Supports using an Ollama instance as an alternative LLM backend, configurable via the CBB_OLLAMA_BASE_URL environment variable, alongside the default LM Studio endpoint.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@cbbstart a new extraction run and show me its status"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
cbb_stage1
抽取段 v6.1 管线(箱 v2.2 契约)。设计见 ../设计/抽取段-v6/,验收见
../设计/抽取段-v6/07-吸收机制总清单-v2-20261020.md。
⚠️ 本包不是自包含的:必须同时带上 设计/抽取段-v6/
src/ 与 tests/ 都靠兄弟目录 ../设计/抽取段-v6/schemas/ 工作:
box/writer.py 建箱时从那里读 19 份 schema,tests/unit/test_* 也从那里读。
只复制 cbb_stage1_v6/ 一个目录过去,建箱会直接抛:
BuildError: box schemas not found: <parent>/设计/抽取段-v6/schemas2026-10-21 实测:只复制 cbb_stage1_v6/ 时 46 个测试失败;把 设计/ 一起复制后
1027 个全过。所以拷贝时保持这个相对位置:
<任意父目录>/
├── cbb_stage1_v6/ # 本包
└── 设计/抽取段-v6/schemas/ # 必需,不是文档Related MCP server: goldenpipe
新机器上跑起来
需要 Python 3.12(pyproject.toml 的 requires-python,preflight.MIN_PYTHON
与它保持一致——两处曾一个写 3.11 一个写 3.12,preflight 会对 pip 装不上的环境报 READY)。
$env:PYTHONUTF8='1' # 中文路径必须;否则 Windows 按 ANSI 代码页读写
# 先看这台机器上有哪个 3.12。`py -3.12` 只在官方安装包下有效;
# uv 装的 Python 会被注册到命名空间里(如 Astral/CPython3.12.14),必须用全名。
py -0 # 列出现有解释器与它们的标识
py -3.12 -m venv .venv # 若上面显示的就是 3.12,用这行
# 否则(uv / 其它命名空间):
# py -V:Astral/CPython3.12.14 -m venv .venv
.venv\Scripts\python -m pip install -e ".[dev]"
.venv\Scripts\python -m pytest -q2026-10-21 在本机实测:py -3.12 报 exit=103(找不到),
py -V:Astral/CPython3.12.14 正常。不要照抄某台机器的绝对路径,先 py -0 看清单。
jieba 首次分词会构建词典缓存(写到 %TEMP%),第一次跑会慢几秒,属正常。
模型端点(都可覆盖)
用途 | 默认 | 覆盖用的环境变量 |
主推理 / 嵌入 |
|
|
Ollama 备选 |
|
|
API key | 空 |
|
judge(异档验证) | 未配置 |
|
默认 mock 引擎,不联网;真实 LLM 只在 --live 下启用。没有模型服务时管线仍能跑完
(推理类能力会记为 unavailable,不伪造结果)。
命令
.venv\Scripts\python -m cbb_stage1.cli lint-box ..\设计\抽取段-v6\example
.venv\Scripts\python -m cbb_stage1.cli run --book <book> --chapters 1-3 --out artifacts\box-3ch
.venv\Scripts\python -m cbb_stage1.cli run --book <book> --chapters 1 --live # 走真实模型
.venv\Scripts\python -m cbb_stage1.cli build-box --work <work> --out <box>
.venv\Scripts\python -m cbb_stage1.cli acceptance acceptance\07.yaml门禁(提交前应全绿)
.venv\Scripts\python -X utf8 -m cbb_stage1.qc.id_sources # ID 单一来源
.venv\Scripts\python -X utf8 -m cbb_stage1.cli wiring # 模块可达
.venv\Scripts\python -m ruff check src\cbb_stage1 --select F401,F811,F841
.venv\Scripts\python -m vulture src\cbb_stage1 .vulture_whitelist.py --min-confidence 80
.venv\Scripts\python -X utf8 tools\call_wiring.py --root src\cbb_stage1 # 公开 API 是否真被调用
.venv\Scripts\python -X utf8 tools\field_coverage.py <box> # 必填字段是否恒空tests/unit/test_portability.py 把「换一台 Windows 11 机器还能跑」这件事也钉成了测试:
源码里不许有绝对路径 / 用户名 / 桌面目录,端点必须可被环境变量覆盖,测试不许绑定本机路径。
换机时的已知依赖
依赖 | 说明 |
中文路径 | 所有 Python 调用带 |
录制夹具 |
|
环境变量夹具 | 别的录制跑可用 |
长路径 | 仓库内最长 121 字符;放到很深的目录时注意 Windows 260 上限 |
This server cannot be deployed
Maintenance
Related MCP Connectors
Bounded tools for rendering, extraction, RAG, enrichment, local discovery and review analysis.
66 public tools: semantic search, queues, inboxes, webhooks, approvals, DNS, images, PDF and more.
Governed data discovery, exact queries, decisions, simulations, and runtime utilities over MCP.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceExposes RAG and document intelligence pipelines as 8 composable tools for MCP-compatible clients, enabling querying, indexing, classifying, extracting, and assessing documents.1MIT
- AlicenseNot gradedqualityFmaintenanceEnables orchestrating data quality checks, transformation, and deduplication pipelines via an MCP interface. Offers tools for listing pipeline stages, validating wiring, running the full check-transform-match pipeline, and explaining configurations.MIT
- AlicenseNot gradedqualityCmaintenanceExposes standard MCP tools for secure calculation, knowledge-base retrieval, document statistics, and sensitive operations like email, export, and deletion with human-in-the-loop approval.MIT
- AlicenseNot gradedqualityBmaintenanceEnables MCP-compatible agents to run the complete document extraction loop as remote tools: upload files, submit extraction jobs with plain-language instructions, answer clarifying questions, inspect review-needed rows and failed pages, and download results as spreadsheets or JSON.MIT