playbooks

home / mcp / paper read mcp server

Paper Read MCP Server

Local MCP server for parsing PDFs, extracting structures, analyzing math, generating code, and visualizing experiments with local LLM support.

Installation
Add the following to your MCP client configuration file.

Configuration

View docs
{
  "mcpServers": {
    "lxy-hqu--mcp-for-paper-read-based-on-ai-ide": {
      "command": "C:\\\\Program Files\\\\nodejs\\\\node.exe",
      "args": [
        "E:\\\\path\\\\to\\\\-mcp-for-paper-read-based-on-AI-IDE\\\\dist\\\\server.js"
      ]
    }
  }
}

You run a local MCP server that analyzes and processes scientific papers on your machine. It provides local PDF parsing, structured extraction, deep math understanding, code generation, and visualization, with optional acceleration from a local large model. This setup helps you build reproducible analysis pipelines, store results in a local database, and generate Markdown reports with correctly rendered math.

How to use

Install and run the local MCP server through your MCP client. You can load a PDF, let the system summarize the paper, extract methods, parse equations, generate PyTorch code, render diagrams, and produce a full Markdown analysis report. All processing happens locally, with optional local model acceleration.

How to install

Prerequisites: Node.js (v16+), Git, and optionally Ollama for local LLM acceleration (default port 11434)

1. Clone the project repository to your workstation

git clone https://github.com/Lxy-hqu/-mcp-for-paper-read-based-on-AI-IDE.git
cd -mcp-for-paper-read-based-on-AI-IDE

2. Install dependencies

npm install

3. Compile the project

npx tsc

4. Configure Trae for MCP access (recommended auto-configuration)

{
  "mcpServers": {
    "local-papers": {
      "command": "C:\\Program Files\\nodejs\\node.exe", 
      "args": [
        "E:\\path\\to\\-mcp-for-paper-read-based-on-AI-IDE\\dist\\server.js"
      ],
      "disabled": false,
      "autoApprove": []
    }
  }
}

Tip: You can find the Node.js path by running the terminal command where node.

Additional setup and usage notes

Once Trae is restarted, the MCP client should show a connected server labeled as local-papers. You can test by asking it to load a file, for example: “请使用 pdf_loader 读取这个文件：E:\my_paper.pdf” (use your own file path).

Usage examples

Math formula parsing, code generation, visualizations, and report creation are available as dedicated capabilities. You can issue natural language prompts such as:

- “解析这篇论文的数学公式，并将符号定义存入数据库。”

- “根据这篇论文的方法部分，生成 PyTorch 模型代码，并提取超参数。”

- “为这篇论文的模型结构生成一个 Mermaid 流程图。”

- “为这篇论文生成一份完整的 Markdown 分析报告。”

Available tools

pdf_loader

Reads PDF files locally and provides raw content and metadata to downstream modules.

structure_parser

Segments and structures the paper into meaningful sections for analysis.

math_explainer

Extracts mathematical formulas, builds ASTs, and stores symbols in the local database.

code_generator

Generates PyTorch model code and training scripts based on the methods section.

visualization

Creates Mermaid diagrams and variable dependency graphs to illustrate models and experiments.

report_generator

Produces a complete Markdown analysis report with summaries, structure, visuals, and code configurations.

summarizer

Generates intelligent summaries and method sections, with optional deep content understanding via a local LLM.