ohhmoo/speqtro

开发工具

本页由 AI 翻译，技术细节请以原文为准。查看原文

22 stars 0 forks 质量 73 趋势 10MIT

Repository Install

SPEQTRO 是一款开源的 CLI 和 MCP 服务器，利用 Claude 解释核磁共振、红外和质谱数据——通过明文查询识别未知化合物并验证结构。

概览

化学家的自主光谱推理剂。 - 跨NMR、MS、IR、结构分析和数据库搜索——配备斜杠命令、光谱学主题界面和对话式分析——speqtro web打开浏览器界面，包含聊天、验证、预测和分析历史——JCAMP-DX、CSV峰值列表、Bruker FID、MestReNova导出和图片——CASCADE（1H、13C预测）、ChefNMR（结构阐明）、SSIN（红外分类）、ICEBERG/ms-pred（MS/MS碎片化）、DP4/DP5（结构验证）——speqtro fetch-weights可从HuggingFace拉取CASCADE、SSIN、DP5和ICEBERG——工具无需机器学习模型，使用加法增量规则和经验表——每个结果都包含方法、数据源和可重复性参数安装后运行交互向导。它保存你的Anthropic API密钥，允许你选择Claude模型（Opus为最佳精度，Haiku为速度），并验证你的环境。

概述（AI 翻译）

化学家的自主光谱推理剂。


	SPEQTRO 将代理人工智能整合进一个CLI工具，帮助化学家从光谱数据中识别未知结构、预测化学位移、搜索参考数据库并生成可重复的分析报告——所有这些都来自终端。

特色

15个内置工具涵盖核磁共振、MS、红外、结构分析和数据库搜索
交互式REPL，带有斜杠命令、光谱学主题界面和对话式分析
Web GUI — speqtro web 打开浏览器界面，包含聊天、验证、预测和分析历史
多格式输入 — JCAMP-DX、CSV 峰值列表、Bruker FID、MestReNova 导出及图片
厂商机器学习模型 — CASCADE（1H，13C预测）、ChefNMR（结构阐明）、SSIN（红外分类）、ICEBERG/ms-pred（MS/MS碎片化）、DP4/DP5（结构验证）
一命令权重下载 — speqtro fetch-weights 从HuggingFace中拉取CASCADE、SSIN、DP5和ICEBERG
启发式后备——工具无需机器学习模型，使用加法增量规则和经验表

本页由 AI 翻译，技术细节请以原文为准。

README

以下为原文 README 节选，完整内容请查看原文。

Autonomous spectroscopy reasoning agent for chemists.


	SPEQTRO wraps Agentic AI into a CLI tool that helps chemists identify unknown structures from spectroscopic data, predict chemical shifts, search reference databases, and generate reproducible analysis reports — all from the terminal.

Features

15 built-in tools across NMR, MS, IR, structure analysis, and database search
Interactive REPL with slash commands, spectroscopy-themed UI, and conversational analysis
Web GUI — speqtro web opens a browser interface with chat, verify, predict, and analysis history
Multi-format input — JCAMP-DX, CSV peak lists, Bruker FID, MestReNova exports, and images
Vendored ML models — CASCADE (1H, 13C prediction), ChefNMR (structure elucidation), SSIN (IR classification), ICEBERG/ms-pred (MS/MS fragmentation), DP4/DP5 (structure verification)
One-command weight download — speqtro fetch-weights pulls CASCADE, SSIN, DP5, and ICEBERG from HuggingFace
Heuristic fallbacks — tools work without ML models using additive increment rules and empirical tables
Mandatory citations — every result includes method, data sources, and parameters for reproducibility

Requirements

Python 3.10 – 3.13
An Anthropic API

Installation

pip (recommended)

pip install speqtro

With optional extras

# With RDKit (structure drawing, SMILES validation)
pip install "speqtro[chemistry]"

# With web GUI
pip install "speqtro[web]"

# With MCP server support
pip install "speqtro[mcp]"

# With all ML backends (PyTorch + TensorFlow)
pip install "speqtro[all]"

pipx (isolated environment, no dependency conflicts)

pipx install speqtro

# With extras
pipx install "speqtro[chemistry]"

Development / editable install

git clone https://github.com/OhhMoo/SPEQTRO-Agent.git
cd SPEQTRO-Agent
pip install -e ".[dev,chemistry]"

Workflow

Step 1 — First-time setup

Run the interactive wizard after installation. It saves your Anthropic API key, lets you choose the Claude model (Opus for best accuracy, Haiku for speed), and verifies your environment.

speqtro setup

Step 2 — Download ML model weights

SPEQTRO ships the vendored model code but not the weights (too large for git). Pull everything from HuggingFace in one command:

speqtro fetch-weights

This downloads CASCADE, SSIN, DP5, and ICEBERG into ~/.speqtro/models/. To download a specific model only:

speqtro fetch-weights --model cascade2    # 13C NMR predictor
speqtro fetch-weights --model iceberg     # MS/MS fragmentation
speqtro fetch-weights --model ssin        # IR functional group classifier
speqtro fetch-weights --model dp5         # Bayesian NMR scoring

在 GitHub 查看完整 README

安装

This server does not publish a one-line install command.

Open the repository installation guide 新手？先看 mcpskill.org 教程 →

配置

{
  "mcpServers": {
    "speqtro": {
      "command": "speqtro",
      "args": ["mcp", "serve", "--stdio"]
    }
  }
}