Skip to main content
Glama
Sign In
Insights Knowledge Base MCP Server

Insights Knowledge Base MCP Server

by v587d
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

Insights Knowledge Base(IKB) MCP Server

>>>中文版

🍭A free, plug-and-play knowledge base. Built-in with 10,000+ high-quality insights reports, packaged as MCP Server, and secure local data storage.

⚠️⚠️ All collected reports in this project come from free resources on official research report websites. ⚠️⚠️

Features

  1. 🍾 Zero configuration required, designed for plug-and-play usage.

  2. 🚀 Built-in Qwen3-Embedding-0.6B embedding model, related reports can be retrieved through vector search.📢 Report details can also be searched via keyword retrieval.

  3. 🍥 over 100 insights reports from well-known consulting firms such as McKinsey, PwC, and BAIN have been collected, including 6,000+ report pages, covering 70+ topics.

  4. 💎 Real-time online browsing of full reports in MCP Client.

  5. 🎉 Ultra-fast response: All Function_call returns typically <1 second, keyword-based queries <150ms.

  6. 🎨 Paste private local documents into the library_files folder (create it manually if absent; name must match). Configure VLM models/parameters in .env (e.g., VLM_MODEL_NAME=qwen2.5-vl-72b-instruct) for local document extraction, parsing, and recognition.

  7. 🦉 Permanently free—no wasted effort collecting reports. Share reliable, copyright-compliant resources via issues.

  8. 🔔 Commit to weekly report updates; bug fixes depend on personal whim (I'm not an engineer 🤭).

Related MCP server: MyAIGist MCP

Optimizations as of June 30

  1. Added 2000+ report pages.

Future Directions

  1. Continuous report updates.

  2. Prompt engineering optimization.

Newest Files Profile

{
    "statistics": {
        "total_files": 174,
        "total_pages": 9320,
        "unique_publishers": 9,
        "unique_topics": 93,
        "last_updated": "2025-06-30T10:08:35.928329"
    },
    "details": {
        "publishers": [
            "",
            "Accenture",
            "BAIN",
            "BCG",
            "CBS",
            "Deloite",
            "McKinsey",
            "PWC",
            "亿欧"
        ],
        "topics": [
            "",
            "AI",
            "AI Agent",
            "Africa",
            "Aftermarket",
            "Asian American",
            "Auto",
            "Aviation",
            "Beauty",
            "Business",
            "Chemical industry",
            "Chemicals",
            "Chinese banking",
            "Chinese securities",
            "Consumer Goods",
            "Decarbonation",
            "Decarbonization",
            "Digital",
            "ESG",
            "Economy",
            "Economy and Trade",
            "Education",
            "Electric two wheelers",
            "Employment",
            "Energy",
            "Europe",
            "FMCG",
            "Fashion",
            "Finance",
            "Financial Technology",
            "Financial service",
            "Fintech",
            "Food-meatless",
            "Gen Z",
            "Global banking",
            "Global energy",
            "Global insurance",
            "Global macroeconomic",
            "Global materials",
            "Global private market",
            "Global private markets",
            "Global trade",
            "Grocery",
            "Grocery retail",
            "Health",
            "Healthcare",
            "Human capital",
            "Hydrogen",
            "Insurance",
            "Investing",
            "Investment management",
            "Labor market",
            "Latinos",
            "Low-altitude Economy",
            "Luxury Goods",
            "Luxury goods",
            "M&A",
            "Maritime",
            "Media",
            "Medical Health",
            "Medtech",
            "Net zero",
            "New Energy Vehicle",
            "New era",
            "Packing",
            "Payments",
            "Pet Food",
            "Population",
            "Power",
            "Private Equity",
            "Private market",
            "Productivity",
            "Quantum",
            "Real estate",
            "Retail",
            "Retail Digitalization",
            "Retailers",
            "Risk",
            "Small business",
            "Smart Home",
            "Smart hospital",
            "Sporting goods",
            "Sustainability",
            "Sustainable",
            "Tax-free",
            "Technology",
            "Travel",
            "Truck",
            "United Kingdom",
            "VSOC",
            "Wealth management",
            "Workplace",
            "连锁经营"
        ]
    }
}

Installation (Beginner-Friendly)

💡Pro tip: Stuck? Drag this page to an LLM client (like DeepSeek) for step-by-step guidance. Actually, these instructions were written by DeepSeek too...

Prerequisites: Python 3.12+ (Download from official website and ADD ENVIRONMENT PATH)

Install UV:

pip install uv

1. Clone the project(Confirm successfully installed Git and Git LFS)

git clone https://github.com/v587d/InsightsLibrary.git
cd InsightsLibrary
git lfs pull

2. Create virtual environment

uv venv .venv  # Create dedicated virtual environment

# Activate environment
# Windows:
.\.venv\Scripts\activate
# Mac/Linux:
source .venv/bin/activate

3. Install core dependencies

uv install .  # Note the trailing dot indicating current directory

4. Create environment variables (for future needs)

notepad .env  # Windows
# Or
nano .env     # Mac/Linux

5. Configure MCP Server

  • VSCode.Cline

Note: Replace <Your Project Root Directory!!!> with actual root directory.

{
  "mcpServers": {
    "ikb-mcp-server": {
      "command": "uv",
      "args": [
        "--directory",
        "<Your Project Root Directory!!!>", 
        "run",
        "ikb_mcp_server.py"
      ]
    }
  }
}

  • Cherry Studio

    • Command: uv

    • Arguments:

--directory
<Your Project Root Directory!!!>
run
ikb_mcp_server.py

Adding Private Documents to ikb_mcp_server

  1. Configure VLM models and parameters in .env:

    VLM_API_KEY=<API Key>
VLM_BASE_URL=<Base URL> # https://openrouter.ai/api/v1
VLM_MODEL_NAME=<Model Name> # qwen/qwen2.5-vl-72b-instruct:free

  2. Upload the PDF document to the library_files folder under the project root directory.

  3. Manually run main.py.

# Navigate to the project root directory
# Activate the virtual environment
uv run main.py
(InsightsLibrary) PS D:\Projects\mcp\InsightsLibrary> uv run main.py
[INFO] extractor: PDF extraction initialized | Files directory: library_files | Pages directory: library_pages
[INFO] extractor: Starting scan of directory: library_files
[INFO] extractor: Found 69 PDF files
[INFO] extractor: Scan completed | Total files: 69 | Processed: 0 | Failed: 0
[INFO] recognizer: No pages to process.
# Data has been updated to the database
============================================================
Confirm if you need to create text vector embeddings
⚠️ This process may take approximately 20 minutes
============================================================
Create embeddings? (Enter Y or N): 
# Y: create text vector embeddings
# N: Skip text vector embeddings and exit program

License

This project is licensed under the MIT License. See the LICENSE file for details.

Optimization Updates as of June 17th

  1. 💡Optimized models.py: Improved data query efficiency by 1,000%

  2. 💡Optimized extractor.py: Slightly enhanced PDF extraction efficiency

  3. 💡Optimized recognizer.py: Boosted image comprehension efficiency by 50%

  4. 💡Optimized ikb_mcp_server.py:

    • Added pagination functionality

    • Displayed local paths of referenced files

  5. 💡Add MIT License(https://github.com/v587d/InsightsLibrary/pull/1#issuecomment-2969226661)

  6. 📦 Overall compressed project package size reduced by approximately 50%

  7. 💡Streamline Private Document Handling

  8. 💡Fixed other identified bugs

Optimizations as of June 22

  1. Added embedder.py: Implements text vectorization indexing via local Qwen3-Embedding-0.6B model, stored in faiss_index.

  2. Modified main.py: Closed-loop workflow PDFExtractor → IMGRecognizer → Embedder (optional).

  3. New @mcp.tool(): get_similar_content_by_rag: Finds most similar document content via vector similarity (RAG).

  4. All admin-uploaded reports now support online viewing → Removed library_files folder to reduce project size.

  5. Added 2000+ report pages.

Available Tools

3 tools
get_similar_content_by_ragA

该方法用于通过计算用户输入与文档内容向量之间相似度，进而找到最相似的文档内容，即RAG。 ⚠️注意：1. 当LLM无法从用户输入中提取明确指令时，优先使用此方法。 2. 当LLM引用该方法返回的结果时，必须用markdown格式明确、醒目告知用户引自哪篇报告和具体访问地址！ 比如“观点引自《21世纪CEO的成功法则》第10、16页。(查看完整报告)[<如果"download_url"不为空，填入download_url>]” ！！！注意每份报告单独列举 download_url，不要笼统指向某一个可能不存在的地址。

 参数：
    user_query: str 必填。

返回：
    page_number: 报告详情页页码
    page_title: 标题
    page_abstract: 摘要
    page_content: 内容
    file_name: 报告名称
    local_path: 报告存放于本地地址
    download_url: 报告网络链接
    published_by: 报告发布机构
    published_date: 报告发布日期
    vector_similarity: 向量相似度

LLM需将该方法返回结果组织成通畅的语言传达给用户。

Inputs: user_query (required).

search_content_detailA

该方法用于查询符合多条件组合的报告详情页面。LLM需根据用户输入的消息(user_message)提炼出以下参数。 ⚠️注意：当LLM引用该方法返回的结果时，必须用markdown格式明确、醒目告知用户引自哪篇报告和具体访问地址！ 比如“观点引自《21世纪CEO的成功法则》第10、16页。(查看完整报告)[<如果"download_url"不为空，填入download_url>]” ！！！注意每份报告单独列举 download_url，不要笼统指向某一个可能不存在的地址。

参数：
    keywords: List[str] = None, 报告详情页的关键词。
    title: str = "", 报告详情页标题包含词。
    content: str = "", 报告详情页内容包含词。
    publisher: str = "", 报告发布者。
    start_date: Optional[datetime] = None, 报告查询开始日期。
    end_date: Optional[datetime] = None, 报告查询结束日期。
    match_logic: str = "OR", 匹配逻辑。"OR" 或者 "AND"，二选一，**优先用 "OR"**。
    page_index: int = 1, 页码，默认仅显示第一页。

返回：
    results：报告详情
      - file_name: 详情页来自于哪份报告名
      - page_number: 页码
      - page_abstract: 摘要
      - page_content: 完整内容
      - page_keywords: 详情页关键词
      - published_by: 报告发布机构
      - published_date:报告发布日期
      - local_path: 报告存放于本地地址
      - download_url: 报告网络链接
      - matched_keywords: 匹配关键词组
    current_page：当前页码。⚠️当前页码小于总页码时，LLM需在结尾处提示用户可输入“下一页”查询更多记录。
    total_pages： 总页数
    total_matches： 总匹配记录条数

LLM需将该方法返回结果组织成通畅的语言传达给用户。

Inputs: title, content, end_date, keywords, +4 more.

search_report_profileA

该方法用于查询多条件组合的报告整体概况。LLM需根据用户输入的消息(user_message)提炼出以下参数。 ️⚠️注意：当LLM引用该方法返回的结果时，必须用markdown格式明确、醒目告知用户引自哪篇报告和具体访问地址！ 比如“观点引自《Open source technology in the age of AI》。(查看完整报告)[<如果"download_url"不为空，填入download_url>]” ！！！注意每份报告单独列举 download_url，不要笼统指向某一个可能不存在的地址。

参数：
    keywords: List[str] = None, 整篇报告的关键词。
    title: str = "", 报告标题包含词。
    content: str = "", 报告内容包含词。
    publisher: str = "", 报告发布者。
    start_date: Optional[datetime] = None, 报告查询开始日期。
    end_date: Optional[datetime] = None, 报告查询结束日期。
    match_logic: str = "OR", 匹配逻辑。"OR" 或者 "AND"，二选一，**优先用 "OR"**。

返回：
    results：报告概览
      - file_name: 报告名称
      - topic: 报告主题
      - content: 报告整体摘要
      - published_by: 发布机构
      - published_date: 发布日期
      - local_path: 报告存放于本地地址
      - download_url: 报告网络链接
      - matched_keywords: 匹配关键词组
    current_page：当前页码。⚠️当前页码小于总页码时，LLM需在结尾处提示用户可输入“下一页”查询更多记录。
    total_pages： 总页数
    total_matches： 总匹配记录条数

LLM需将该方法返回结果组织成通畅的语言传达给用户。

Inputs: title, content, end_date, keywords, +4 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Deploy Server
    One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

    TDQS

    A4/5.0

    Scored across 3 tools

    Disambiguation5/5

    Each tool targets a distinct retrieval operation: report overviews (search_report_profile), page-level details (search_content_detail), and semantic similarity (get_similar_content_by_rag). There is no ambiguity in their purposes.

    Naming Consistency4/5

    All tool names use snake_case, with two starting with 'search_' and one with 'get_'. While the 'get_' prefix differs, the pattern is still predictable and descriptive. Minor inconsistency in verb choice.

    Tool Count4/5

    Three tools is appropriate for a focused knowledge base retrieval server. It covers the core query types without being excessive or insufficient for the stated purpose.

    Completeness4/5

    The tool set covers structured search by report profile and content, plus semantic search. For a read-only knowledge base, this is reasonably complete. Missing write operations are acceptable given the domain.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Resources

    Related MCP Connectors

    Related MCP Servers

    • MCP Code Analysis Server

      johannhartmann
      A
      license
      Not graded
      quality
      D
      maintenance
      An intelligent server that provides semantic code search, domain-driven analysis, and advanced code understanding for large codebases using LLMs and vector embeddings.
      10
      MIT

    • MyAIGist MCP

      schwim23
      F
      license
      Not graded
      quality
      D
      maintenance
      A local document intelligence and knowledge management server for Claude Desktop that provides RAG-powered Q\&A, media transcription, and URL crawling. It features 11 tools for processing various file types and managing a persistent local vector store with zero infrastructure costs.
      1
      -