clawsgo-science-mcp
Provides optional LLM-powered enhancements for paper generation, review, and ideation via OpenAI-compatible endpoints, falling back to local templates when no API key is configured.
Click on "Install 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., "@clawsgo-science-mcpReproduce the paper at ~/paper.pdf and show me the deliverables."
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.
clawsgo-science-mcp
ClawsGO Science 复刻 — 一条 MCP 从「构思」到「论文交付」的全流程编程工作台。
一个完全本地自建的 MCP server,提供论文领域的端到端自动化:
🧪 复现线 — 拿到一篇论文 PDF,五步闭环复现:解析 → 方案 → 生成代码 → 沙箱运行比对 → 产出交付物
✍️ 写作线 — 按章节引导生成论文内容,导出 LaTeX / PDF / DOCX
🔬 研究线 — 构思 → 选题 → 多视角假设辩论 → 实验设计 → 结果自旋门决策
📦 交付线 — 把复现/写作的全部实物(源码/数据/报告/图/导出物)分门别类交付
特色:免 API key。 文献检索用 OpenAlex 免鉴权接口(覆盖 arXiv 预印本),LLM 能力为可选增强——没有 LLM 时自动回退到本地模板,整条链路仍然可用。
独立自建、仅用公开材料,不附属于、不依赖 ClawsGO 官方站点或 API。
功能总览
工具 | 所属线 | 作用 |
| 复现 | 解析 PDF → 复现方案 → 生成代码 → 沙箱运行比对 → 交付物 |
| 复现 | 查询复现任务的异步状态与阶段进度 |
| 写作 | 按章节引导撰写(abstract/problem/…/results/references) |
| 写作 | 将已写 doc 导出为 LaTeX / PDF / DOCX |
| 研究 | 构思→选题→研究缺口→候选假设→多视角评审→实验计划 |
| 研究 | 把复现的真实实验结果并入论文对应章节 |
| 研究 | 结果自旋门:PROCEED / REFINE / PIVOT 决策建议 |
| 交付 | 列出任务(复现/写作)的交付物清单,按类型分类 |
全链路:从构思到论文
ideate_paper(构思/假设/实验计划)
│
▼
reproduce_paper ──► reproduce_status(异步轮询,五步闭环)
│
├────────────────────────────┐
▼ ▼
research_verdict(PROCEED/REFINE/PIVOT) write_section(撰写各章节)
│ │
▼ ▼
inject_results(实验数据并入 results) export_document(LaTeX/PDF/DOCX)
│ │
└──────────────► get_deliverables(交付全部实物)Related MCP server: research-automation-mcp-server
快速开始
安装
要求 Python ≥ 3.10。
git clone https://github.com/liudongyan13701205717-source/clawsgo-science-mcp.git
cd clawsgo-science-mcp
pip install -e . # 最小安装(只带 MCP 本体)
# 需要复现线时,装上运行依赖(推荐):
pip install -e ".[reproduce,dev]"依赖说明:
依赖 | 用途 | 必需? |
| MCP 协议与 FastMCP 运行 | ✅ 必需(base) |
| 论文 PDF 解析 | 复现线 |
| 沙箱执行复现代码、绘制收敛图 | 复现线 |
本机 | LaTeX→PDF 导出 | 可选(缺省时自动用内置 PyMuPDF 渲染出 PDF) |
验证安装:
python -m clawsgo_self.server --help 2>&1 | Out-Null # 能启动即安装成功(stdio server)
python -m pytest tests/ -q # 跑内置测试,应全绿在 opencode / Claude Code 中配置
把 server 注册为本地 MCP(以 opencode 为例,~/.config/opencode/opencode.json):
{
"mcp": {
"clawsgo-self": {
"type": "local",
"command": ["python", "-m", "clawsgo_self.server"],
"cwd": "<你的工作区路径>",
"enabled": true,
"environment": {
"PYTHONPATH": "<你的工作区路径>"
}
}
}
}用绝对路径指向解释器与工作区,避免命中系统自带的 Python Store 占位符。
重启客户端后即可看到全部 8 个工具。
真实使用示例
1. 让 MCP 写一篇论文
paper_id = "demo"
write_section(paper_id, "abstract", "为《边缘设备上的轻量推理》写摘要,主题=推理速度优化", "markdown")
write_section(paper_id, "problem", "问题定义:边缘设备推理延迟与能耗瓶颈", "markdown")
write_section(paper_id, "modeling", "建模:轻量化网络与量化方案", "markdown")
write_section(paper_id, "solution", "求解:剪枝/蒸馏/量化组合策略", "markdown")
write_section(paper_id, "results", "实验设置与评估指标", "markdown")
write_section(paper_id, "references", "参考文献", "markdown")
export_document(paper_id, "pdf") # 产出 demo/doc.pdf + doc.tex + doc.html2. 构思一个研究课题
ideate_paper("大语言模型的轻量化可解释方法", "my_proj")
# 返回:研究缺口、候选假设、多视角评审(novelty/rigor/feasibility 加权)、实验计划3. 复现一篇论文并注入结果
tid = reproduce_paper("path/to/paper.pdf")["task_id"]
# 轮询 reproduce_status(tid) 直到 done
research_verdict(tid) # 根据实验结果给出 PROCEED/REFINE/PIVOT
inject_results("my_proj", tid) # 把真实指标表 + 收敛图写进 my_proj 的 results 章节
get_deliverables(tid) # 拿到复现的全部交付物架构
clawsgo_self/
├── server.py # MCP stdio server,注册全部 8 个工具
├── core/ # 布局/存储/Layout + 可选 LLM 连接层(无 key 会回退模板)
├── parse/ # 论文 PDF 解析(PyMuPDF)
├── reproduce/ # 五步复现闭环:tasks/codegen/harness/pipeline
│ ├── codegen.py # 提取超参 + 生成 numpy/torch 复现代码
│ ├── harness.py # 沙箱执行 + 自愈重试 + 指标/图采集
│ └── pipeline.py # 编排五步,产出 results.json / plan.json / deliverables/
├── write/ # 写作:doc(DocStore)/templates/validate
├── export/ # 导出:md→latex/html/docx + PDF 渲染
├── research/ # 研究线:lit(OpenAlex 免 key 检索)/ideate/hypoth/design/inject
└── deliver/ # 交付:get_deliverables(源码/数据/报告/图 分类)存储布局
.clawsgo-self/ # 运行产物,已 gitignore
├── env # 本地配置(可选 LLM 端点),不提交
├── projects/{paper_id}/ # 写作项目:doc.md / doc.pdf / doc.tex / sections/
└── tasks/{task_id}/ # 复现任务:parse/ code/ runs/ deliverables/可选增强:LLM
在 .clawsgo-self/env 写入 OpenAI 兼容端点即可启用 LLM 增强(更自然的章节生成/评审/构思):
BASE_URL=https://.../v1
API_KEY=sk-...
MODEL=...未配置时,所有线自动回退到本地模板,功能不受阻断。
测试
pip install -e ".[dev]"
python -m pytest tests/ -q # 40 项,全离线可跑(文献/复现用模拟数据)测试覆盖四条线上的单元 + stdio 端到端(真实 spawn MCP server 并调用工具)。
反馈与贡献
遇到 bug、功能建议或有任何问题,欢迎通过 GitHub Issues 反馈:
👉 https://github.com/liudongyan13701205717-source/clawsgo-science-mcp/issues
🐛 Bug:请附上复现步骤、报错信息(含
File "...", line ...堆栈)与相关paper_id/task_id。💡 Feature:说明你的使用场景与期望行为。
🤝 Contribute:Fork 后提 PR 即可;请先运行
python -m pytest tests/ -q确保全绿。
License
本项目为独立开发的教育/研究工作,与 ClawsGO 及其商标无关联。
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Connectors
Search arXiv/Semantic Scholar/OpenAlex + medical evidence (PubMed/Europe PMC) + LaTeX/PDF tools.
Research intelligence for AI coding agents. 2M+ CS papers with evidence and tradeoffs.
Autonomous research agent that pays every source it cites in USDC on Arc via x402 micropayments.
Self-hosted AI-native knowledge workspace with hybrid search, GraphRAG, and MCP.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables local analysis of scientific papers including PDF parsing, mathematical formula extraction with AST generation, PyTorch code generation from methodology, and automated Markdown report generation with visualizations.15
- AlicenseNot gradedqualityDmaintenanceEnables autonomous research by integrating multiple free sources including web search, Wikipedia, arXiv, and Crossref, with no API keys required.MIT
- FlicenseNot gradedqualityCmaintenanceEnables LLMs to autonomously research the web, store knowledge in a local vector database, and retrieve semantic memory without vendor lock-in.
- FlicenseNot gradedqualityAmaintenanceEnables to turn a research folder into a submission-ready academic paper, with stages for screening, venue selection, drafting, humanization, and final QA. Runs locally without uploading any data.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/liudongyan13701205717-source/clawsgo-science-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server