pydantic-docs
A Model Context Protocol (MCP) server providing local-first access to Pydantic and Pydantic AI documentation with BM25-powered full-text search.
README
Pydantic Documentation MCP Server
A Model Context Protocol (MCP) server providing local-first access to Pydantic and Pydantic AI documentation with BM25-powered full-text search.
Features
- Local-first architecture - Offline-only mode by default
- BM25 full-text search - Fast semantic search across all docs
- Git-based extraction - Direct from source repositories (no HTML scraping)
- Pre-processed data - JSONL files included for instant setup
- Auto-initialization - Builds indices automatically on first run
- Complete coverage - Pydantic v2 and Pydantic AI documentation
Requirements
- Python 3.12+
- uv package manager
- ~15MB disk space (with indices)
Quick Start
# Clone and install
git clone <repository-url>
cd mcp_pydantic_docs
uv sync
# Server auto-builds indices on first run
uv run mcp-pydantic-docs
MCP Client Configuration
Add to your MCP settings (e.g., cline_mcp_settings.json):
{
"mcpServers": {
"pydantic-docs": {
"command": "uv",
"args": [
"--directory",
"/absolute/path/to/mcp_pydantic_docs",
"run",
"mcp-pydantic-docs"
]
}
}
}
Architecture
How It Works
- Source Extraction (
source_extractor.py) - Clones Pydantic repos, extracts documentation from markdown/docstrings → JSONL - Index Building (
indexer.py) - Processes JSONL files → BM25 search indices - MCP Server (
mcp.py) - Serves documentation via MCP tools - Shared Utilities (
utils.py) - HTML/text processing, normalization
Directory Structure
mcp_pydantic_docs/
├── mcp_pydantic_docs/ # Source code
│ ├── mcp.py # MCP server
│ ├── source_extractor.py # Git-based doc extraction
│ ├── indexer.py # BM25 index builder
│ ├── utils.py # Shared utilities
│ └── setup.py # Setup CLI
├── data/ # Search data
│ ├── pydantic.jsonl # Pydantic docs (2.9MB, in git)
│ ├── pydantic_ai.jsonl # Pydantic AI docs (3.3MB, in git)
│ ├── *_bm25.pkl # BM25 index (generated)
│ └── *_records.pkl # Document records (generated)
└── docs_raw/ # Source repos (not in git)
├── pydantic/ # Cloned from GitHub
└── pydantic_ai/ # Cloned from GitHub
Data Flow
GitHub Repos → source_extractor.py → JSONL files → indexer.py → BM25 indices → mcp.py → MCP Client
Available Tools
Search & Retrieval
pydantic_search(query, k=10)- Full-text search with BM25 rankingpydantic_get(path_or_url, max_chars=None)- Fetch full documentation pagepydantic_section(path_or_url, anchor)- Extract specific sectionpydantic_api(symbol, anchor=None)- Jump to API documentation
Health & Admin
health_ping()- Server health checkhealth_validate()- Validate search indicespydantic_mode()- Server configurationadmin_cache_status()- Detailed cache statusadmin_rebuild_indices()- Rebuild search indices
Updating Documentation
Rebuild from Existing JSONL
uv run python -m mcp_pydantic_docs.indexer
Extract Fresh Documentation
# Check status
uv run python -m mcp_pydantic_docs.setup --status
# Download and extract from GitHub
uv run python -m mcp_pydantic_docs.setup --download --build-index
# Clean cache
uv run python -m mcp_pydantic_docs.setup --clean
Configuration
Environment Variables
PDA_DOC_ROOT- Pydantic v2 source pathPDA_DOC_ROOT_AI- Pydantic AI source pathPDA_DATA_DIR- Data directory path
Offline Mode
Default: Enabled (OFFLINE_ONLY = True in mcp.py)
- Blocks remote requests
- Validates file paths
- All content from local cache
Development
Run Tests
uv run pytest
Code Quality
uv run black mcp_pydantic_docs/ # Format
uv run ruff check . # Lint
uv run mypy mcp_pydantic_docs/ # Type check
Troubleshooting
Search indices not found:
uv run python -m mcp_pydantic_docs.indexer
Wrong Python version:
uv python install 3.12
Server won't start:
# Test standalone
uv run mcp-pydantic-docs
# Check indices
uv run python -m mcp_pydantic_docs.setup --status
License
MIT License - see LICENSE file.
Contributing
See CONTRIBUTING.md for:
- Development setup
- Code style
- Testing requirements
- Pull request process
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。