documentation-search-enhanced 1.6.1

Enhanced MCP server for searching documentation with OSINT vulnerability scanning, security analysis, and AWS-style deployment

Documentation Search MCP Server

This Model Context Protocol server delivers documentation search, vulnerability auditing, and project bootstrapping in one place. It runs as a long-lived process that serves requests from MCP-compatible clients such as Claude Desktop or Cursor.

Core Capabilities

• AI-Powered Semantic Search: Vector embeddings + hybrid reranking across 100+ documentation sources for superior relevance.
• Security-First Approach: Scan local Python projects for dependency vulnerabilities with multi-tool analysis.
• Project Scaffolding: Generate production-ready starters (FastAPI, React) and developer environment files.
• Developer Productivity: Learning paths, curated code examples, and library security comparisons on demand.

Quick Start

# Requires Python 3.12+
uvx documentation-search-enhanced@latest

Configure your assistant to launch the server:

{
  "mcpServers": {
    "documentation-search-enhanced": {
      "command": "uvx",
      "args": ["documentation-search-enhanced@latest"]
    }
  }
}

No API key is required. Optionally set SERPER_API_KEY to use Serper-powered search. Without it, the server uses a prebuilt docs index (auto-downloaded from GitHub Releases on startup) plus on-site docs search (MkDocs/Sphinx indexes when available). Control the download with DOCS_SITE_INDEX_AUTO_DOWNLOAD, DOCS_SITE_INDEX_MAX_AGE_HOURS, DOCS_SITE_INDEX_URL(S), and DOCS_SITE_INDEX_PATH. Set server_config.features.real_time_search=false to avoid any live crawling and rely only on the downloaded index. The process stays running and listens for JSON-RPC calls; stop it with Ctrl+C when finished.

Codex CLI

Add the server using Codex’s built-in MCP manager:

codex mcp add documentation-search-enhanced \
  -- uvx documentation-search-enhanced@latest

To run from a local checkout instead:

codex mcp add documentation-search-enhanced \
  -- bash -lc 'cd /path/to/documentation-search-mcp && uv run python -m documentation_search_enhanced.main'

Development Workflow

git clone https://github.com/anton-prosterity/documentation-search-mcp.git
cd documentation-search-mcp
uv sync --all-extras --all-groups  # include dev tools
# Optional: enable Serper-powered search
# echo "SERPER_API_KEY=your_key_here" > .env
uv run python -m documentation_search_enhanced.main

• Run core tests: uv run pytest --ignore=pytest-test-project.
• Run example FastAPI tests: cd pytest-test-project && uv run --all-extras python -m pytest -q.
• Lint: uv run ruff check src. Format: uv run black src (use --check to verify).
• Build distributions via uv build; publish_to_pypi.sh wraps the release flow.

Configuration

Ask your assistant for the current configuration via the get_current_config tool, save it as config.json, then adjust sources or caching preferences. Validate changes locally with uv run python src/documentation_search_enhanced/config_validator.py. Keep secrets in .env rather than committing them.

AI-Powered Semantic Search

The server now features vector-based semantic search with hybrid reranking for significantly improved relevance:

How It Works

• Vector Embeddings: Uses sentence-transformers (all-MiniLM-L6-v2) to generate 384-dimensional semantic embeddings
• Hybrid Scoring: Combines three signals with configurable weights:
  • Semantic similarity (50%): True meaning-based matching via cosine similarity
  • Keyword relevance (30%): Precise term matching for specific queries
  • Source authority (20%): Official docs, code examples, and URL quality signals
• FAISS Index: Lightning-fast similarity search over large documentation corpora

Benefits

• Better Understanding: Finds semantically related content even with different terminology
• Improved Ranking: Hybrid approach balances semantic understanding with keyword precision
• Production Ready: Compact 120MB model, efficient L2-normalized embeddings
• Flexible: Enable/disable vector reranking with use_vector_rerank parameter

Usage

Vector reranking is enabled by default in semantic_search. To disable:

semantic_search(query="FastAPI auth", libraries=["fastapi"], use_vector_rerank=False)

This feature addresses the #1 competitive gap vs Context7 while maintaining our security and scaffolding advantages.

Tools at a Glance

Key MCP tools include get_docs, semantic_search, get_learning_path, get_code_examples, scan_project_dependencies, generate_project_starter, manage_dev_environment, get_security_summary, and compare_library_security.

Contributing & License

Start with the contributor guide in AGENTS.md plus the workflow details in CONTRIBUTING.md. Follow Conventional Commits, document validation steps in pull requests, and update CHANGELOG.md for user-facing adjustments. Released under the MIT License—see LICENSE for the full text.