linked-docs
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Linked Documentation System
A production-ready RAG system with dual interfaces: REST API + MCP Protocol
Enables web applications and AI assistants to intelligently search and reference documentation using hybrid semantic + keyword search.
šÆ What Is This?
This is a complete RAG (Retrieval Augmented Generation) system that provides two ways to access powerful documentation search:
REST API Server (
main.py) - FastAPI-based HTTP server for web applications and integrationsMCP Server (
mcp_server.py) - Model Context Protocol server for AI assistants (Cursor, Claude Desktop)
Both servers share the same hybrid search engine, enabling accurate documentation retrieval whether you're building a web app or empowering an AI assistant.
Key Features
Intelligent Hybrid Search: Combines semantic understanding (FAISS embeddings) with keyword matching (BM25)
Smart Ranking: Title/metadata boosting, multi-chunk document expansion, relevance scoring
Multi-Format Support: PDF, Markdown, and web documentation (via built-in scraper)
MCP Native: Works seamlessly in Cursor, Claude Desktop, and other MCP-compatible tools
Enterprise Ready: Access control, audit logging, local-first architecture
Fast: <200ms search latency, optimized chunking and indexing
Zero Cost: Runs 100% locally, no API keys or cloud dependencies
Related MCP server: RAG In A Box MCP Server
šļø How It Works
āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā
ā AI Assistant (Cursor/Claude) ā
āāāāāāāāāāāāāāāāāāāāāā¬āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā
ā MCP Protocol (JSON-RPC over stdio)
ā¼
āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā
ā mcp_server.py ā
ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā
ā ā Tools: search_documentation(), list_sources() ā ā
ā āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā ā
āāāāāāāāāāāāāāāāāāāāāā¬āāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāāā
ā
āāāāāāāāāāāāāāāāāā¼āāāāāāāāāāāāāāāāā
ā¼ ā¼ ā¼
āāāāāāāāāāā āāāāāāāāāāāā āāāāāāāāāāāāāāāā
ā Hybrid ā ā Access ā ā Audit Logger ā
ā Search ā ā Control ā ā ā
āāāāāā¬āāāāā āāāāāāāāāāāā āāāāāāāāāāāāāāāā
ā
āā Semantic Search (FAISS + embeddings)
ā ⢠Title/metadata boosting
ā ⢠Multi-chunk document expansion
ā
āā Keyword Search (BM25)
⢠Exact term matching
⢠Traditional rankingProject Structure
LinkedDocsMCP/
āāā mcp_server.py # Main MCP server (stdio interface)
āāā main.py # FastAPI server (for testing/debugging)
āāā download_docs.py # Web documentation scraper CLI
āāā connectors/ # Document format handlers
ā āāā pdf.py # PDF extraction
ā āāā markdown.py # Markdown parsing
āāā indexing/ # Search engine core
ā āāā chunker.py # Semantic text chunking
ā āāā embedder.py # Sentence transformers
ā āāā vector_store.py # FAISS vector database
ā āāā keyword_search.py # BM25 implementation
ā āāā hybrid_search.py # Combined search with boosting
āāā schemas/ # Data models
ā āāā config.py # Settings & configuration
ā āāā document.py # Document schemas
āāā core/ # Cross-cutting concerns
ā āāā access_control.py # Permission system
ā āāā audit.py # Query logging
āāā data/ # Local storage
āāā sources/ # Your documents (PDF, MD)
āāā vector_store/ # Indexed vectors & metadataš Quick Start (5 Minutes)
Prerequisites
Python 3.10+
Cursor or Claude Desktop (for MCP integration)
1. Install Dependencies
pip install -r requirements.txtFirst run: Downloads ~80MB embedding model (one-time)
2. Add Documentation
Option A: Download from web
# Download Factorio wiki (example)
python download_docs.py https://wiki.factorio.com/Tutorials --crawl --max 20Option B: Add your own files
# Copy PDFs or Markdown files
copy your-docs.pdf data/sources/
copy your-guide.md data/sources/3. Set Up MCP in Cursor (or other LLM service)
Add to your Cursor MCP config (~/.cursor/mcp.json or C:\Users\<USER>\.cursor\mcp.json):
{
"mcpServers": {
"linked-docs": {
"command": "python",
"args": ["C:/full/path/to/LinkedDocsMCP/mcp_server.py"]
}
}
}4. Restart Cursor & Use!
In Cursor's chat:
What are the different enemy types in Factorio?The AI will automatically search your documentation and provide accurate, cited answers! āØ
Key Features Explained
Hybrid Search
Combines two complementary approaches:
Semantic Search (70% weight)
Uses
sentence-transformers(all-MiniLM-L6-v2 model)Understands meaning: "authentication setup" matches "configuring auth"
Converts text to 384-dimensional vectors
Fast similarity search with FAISS
Keyword Search (30% weight)
Uses BM25 algorithm (same as Elasticsearch)
Exact term matching: great for technical terms, code, etc.
Traditional ranking with document length normalization
Smart Ranking Enhancements:
Title Boosting: Documents whose titles match the query get 3x boost
Multi-Chunk Expansion: Returns up to 3 sequential chunks from highly relevant documents
Document Grouping: Results grouped by source document for better context
Semantic Chunking
Unlike naive character-splitting, this uses smart boundaries:
Markdown headers (
##,###) - keeps sections togetherParagraph breaks (
\n\n) - maintains topical coherenceSentences - fallback for unstructured text
Settings:
Chunk size: 2048 characters (whole sections, not fragments)
Overlap: 200 characters (prevents context loss at boundaries)
Web Documentation Scraper
Built-in tool to download and convert web docs:
# Download single page
python download_docs.py https://wiki.example.com/Guide
# Crawl multiple pages (with smart duplicate detection)
python download_docs.py https://wiki.example.com/Main --crawl --max 50
# Force re-download (skip existing detection)
python download_docs.py https://wiki.example.com/Main --crawl --force
# Filter by language
python download_docs.py https://wiki.example.com/Main --crawl --languages en,deFeatures:
Auto-detects and skips existing pages (saves time & bandwidth)
Respects same-domain and link patterns
Polite crawling with configurable delays
Converts HTML to clean Markdown with metadata
Preserves document structure (headers, lists, tables)
š Security & Access Control
Built-in features:
4-tier access hierarchy: PUBLIC ā INTERNAL ā RESTRICTED ā CONFIDENTIAL
Query-time filtering based on user permissions
Full audit logging (every query tracked)
Local-only processing (no data leaves your machine)
Audit logs (data/audit.log):
{
"timestamp": "2025-10-21T14:30:00Z",
"event_type": "search",
"user_id": "mcp_client",
"query": "enemy types",
"results_count": 5,
"search_time_ms": 143
}āļø Configuration
Edit schemas/config.py or set environment variables:
# Search weights
SEMANTIC_WEIGHT = 0.7 # Meaning-based search
KEYWORD_WEIGHT = 0.3 # Exact term matching
# Chunking
CHUNK_SIZE = 1280 # Larger chunks for complete sections
CHUNK_OVERLAP = 128 # Overlap for context continuity
# Model
EMBEDDING_MODEL = "all-MiniLM-L6-v2" # Fast, accurate, smallš§ Advanced Usage
REST API (for testing/debugging)
# Start FastAPI server
python main.py
# Search via REST
curl -X POST http://localhost:8000/api/v1/search_docs \
-H "Content-Type: application/json" \
-d '{"query": "getting started", "top_k": 5}'
# Interactive API docs
open http://localhost:8000/docsProgrammatic Usage
from indexing.embedder import Embedder
from indexing.vector_store import VectorStore
from indexing.hybrid_search import HybridSearchEngine
# Initialize
embedder = Embedder(model_name="all-MiniLM-L6-v2")
vector_store = VectorStore(embedding_dim=384)
search_engine = HybridSearchEngine(vector_store, keyword_searcher, embedder)
# Search
results = search_engine.search("how to configure auth", top_k=5)
for chunk, score in results:
print(f"{score:.3f}: {chunk.text[:100]}...")Technical Highlights
Hybrid search outperforms pure semantic or keyword alone
Smart chunking preserves document structure
Title boosting dramatically improves ranking quality
Multi-chunk expansion provides complete context
Zero cloud dependencies - privacy-first architecture
MCP native - works with any compatible AI assistant
š Acknowledgments
Built with:
FastAPI - Modern Python web framework
Sentence Transformers - Semantic embeddings
FAISS - Vector similarity search
BM25 (rank-bm25) - Keyword ranking
MCP - Model Context Protocol by Anthropic
Status: ā Demo Ready | Version: 1.0.0 | Updated: October 2025
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelogā¦
Related MCP Connectors
Search your knowledge bases from any AI assistant using hybrid RAG.
Agentic search over your Dewey document collections from any MCP-compatible client.
Team docs served to AI agents over MCP - search, Markdown reads, version pinning, read audit.
Versioned documentation registry and semantic search for AI tools and coding assistants.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceAn MCP server implementation that provides tools for retrieving and processing documentation through vector search, enabling AI assistants to augment their responses with relevant documentation context22 npm263MIT
- FlicenseNot gradedqualityBmaintenanceEnables any MCP-compatible AI assistant to search, filter, and retrieve information from a local document collection using a hybrid search pipeline with vector, BM25, reranking, and LLM enrichment.4-
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to access up-to-date documentation by indexing GitHub repositories and official docs, providing semantic search through MCP.24 npm7MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to query documents in a Bedrock Knowledge Base through the MCP protocol, with tools for semantic search and agentic retrieval.MIT