Verbatim MCP
让 AI agent 直接驱动 Verbatim —— 本机自托管的音视频转写 + 博主观点分析工具。
装好之后,你可以直接对 Claude 说:
把这个 B 站链接转写了,转完告诉我讲了什么
搜一下我的转写库里提到「量化交易」的内容
分析一下这个 UP 主的观点,最多看 20 个视频
Claude 会自己调用转写、轮询进度、取回结果。
前置条件
- Verbatim 本体在跑(默认
http://127.0.0.1:5001)。这个 MCP server 只是个客户端,不含转写能力。 - Python 3.10+(Verbatim 本体锁在 3.9,这里是独立进程,互不影响)。
安装
git clone https://github.com/xyzxinlu-max/getAudio.git
cd getAudio/verbatim-mcp
python3 -m venv .venv && .venv/bin/pip install -e .
接到 Claude Code
claude mcp add verbatim --scope user -- /绝对路径/verbatim-mcp/.venv/bin/verbatim-mcp
接到 Claude Desktop 的话,编辑 claude_desktop_config.json:
{
"mcpServers": {
"verbatim": {
"command": "/绝对路径/verbatim-mcp/.venv/bin/verbatim-mcp"
}
}
}
Verbatim 不在默认地址时,加一个环境变量:
{
"mcpServers": {
"verbatim": {
"command": "/绝对路径/verbatim-mcp/.venv/bin/verbatim-mcp",
"env": { "VERBATIM_URL": "http://127.0.0.1:5002" }
}
}
}
提供的工具
| 工具 | 作用 |
|---|---|
transcribe_urls |
提交视频链接转写(支持播放列表/合集/频道,可指定时间段) |
transcribe_file |
转写本机文件(走软链,不拷贝、不上传) |
check_task |
查转写任务状态 |
get_transcript |
取转写正文 + AI 标题/摘要/标签 |
search_transcripts |
全文搜索转写库 |
list_transcripts |
列出转写库条目 |
analyze_creator |
对一个博主做系统性观点分析(重量级长任务) |
check_analysis |
查分析进度 |
list_analysis_files |
列出分析产出的文档 |
get_analysis_file |
读取某份分析文档 |
verbatim_status |
检查 Verbatim 是否在跑、各引擎 key 配置情况 |
设计说明
转写和分析都是长任务,所以工具是「提交 → 拿 id → 轮询」两段式,不会让 agent 阻塞几分钟干等。每个提交类工具的返回值里都写了下一步该调谁。
纯 HTTP 客户端,不 import Verbatim 任何代码。好处是两边的 Python 版本互不牵扯,Verbatim 挂了也只是工具调用失败,不会拖垮 agent 会话。
长文本默认截断(转写 4000 字、分析文档 8000 字),避免几万字正文无脑塞进上下文。需要全文时传 full=True。
隐私提醒
转写引擎选 gemini / dashscope 时,音频会上传到对应厂商。要完全本地处理就用 whisper(默认),不需要任何 API key。
License
MIT
Metadata
Release files for verbatim-transcribe-mcp 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| verbatim_transcribe_mcp-0.1.0.tar.gz | 7.4 kB | Without provenance | Details for verbatim_transcribe_mcp-0.1.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| verbatim_transcribe_mcp-0.1.0-py3-none-any.whl | Python 3 | none | any | Details for verbatim_transcribe_mcp-0.1.0-py3-none-any.whl |
Total release size: 15.8 kB
Release files / verbatim_transcribe_mcp-0.1.0.tar.gz
| Download URL | verbatim_transcribe_mcp-0.1.0.tar.gz |
|---|---|
| Size | 7.4 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
7cfeb3215a901b0dfd69b669b79b90422bbe901f5a820f34880bd5250d503129
|
|
BLAKE2b-256 checksum How to use checksums |
836a053091ed8b68026fb9ea376c1975a36fed238bd2d4689c11d3f7e08757af
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.14.6
|
Release files / verbatim_transcribe_mcp-0.1.0-py3-none-any.whl
| Download URL | verbatim_transcribe_mcp-0.1.0-py3-none-any.whl |
|---|---|
| Size | 8.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
c3abab4c583ddb06bc03a54cfe7bfeca5f29459a3ce9e7bf2df011376e39a87b
|
|
BLAKE2b-256 checksum How to use checksums |
b6bc5ccae5e7fa6f9aed25c6259389ab571748e289f01c2d945b17eca2d8d98d
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.14.6
|
Security for verbatim-transcribe-mcp 0.1.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.