mcp-library-docs
Provides AI assistants with access to design documentation across multiple code libraries, enabling discovery of existing utilities and patterns to avoid reimplementation.
README
mcp-library-docs
An MCP server that gives Claude access to design documentation across multiple code libraries/repositories. This enables Claude to discover existing utilities, patterns, and conventions - preventing reimplementation of existing functionality.
For tips on how I use this server as part of a broader AI-assisted development workflow, see AI Dev Workflow Tips.
Installation
pip install mcp-library-docs
Quick Start
Zero config required - just create a designs/INDEX.md in your project and the server will auto-discover it.
- Create
designs/INDEX.mdin your project:
# my-project
> Brief description of what this project provides
## Modules
### utils
Helper utilities for common operations.
Key exports: `retry`, `parse_date`, `format_currency`
→ Full doc: utils.md
### api
REST API client and authentication.
Key exports: `ApiClient`, `authenticate`
→ Full doc: api.md
-
Add the MCP server to your AI coding assistant (see Setup below)
-
Claude will now discover your project's design docs when working in that directory.
Setup
Claude Code
Add the server using the CLI:
claude mcp add library-docs -- python -m mcp_library_docs
Or manually add to your MCP config (~/.claude/settings.json or project .mcp.json):
{
"mcpServers": {
"library-docs": {
"command": "python",
"args": ["-m", "mcp_library_docs"]
}
}
}
Cursor
Add to your Cursor MCP settings (.cursor/mcp.json in your project or global config):
{
"mcpServers": {
"library-docs": {
"command": "python",
"args": ["-m", "mcp_library_docs"]
}
}
}
Or use an absolute path to the Python interpreter if needed:
{
"mcpServers": {
"library-docs": {
"command": "/path/to/venv/bin/python",
"args": ["-m", "mcp_library_docs"]
}
}
}
How It Works
When Claude calls list_libraries, the server:
- Auto-discovers current project: Walks up from Claude's working directory to find the nearest
designs/INDEX.md - Loads external libraries: Reads any libraries configured in
config.yaml(see Config File Location) - Returns combined index: All INDEX.md contents with type tags (
[current project],[library],[project])
Claude can then call get_design_doc(library, topic) to read detailed documentation.
Configuration
Configuration is optional. Without it, the server just discovers the current project.
Config File Location
The server searches for config.yaml in platform-specific locations:
| Platform | Primary Location | Fallback |
|---|---|---|
| Linux | ~/.config/mcp-library-docs/ |
- |
| macOS | ~/.config/mcp-library-docs/ |
~/Library/Application Support/mcp-library-docs/ |
| Windows | %APPDATA%\mcp-library-docs\ |
~/.config/mcp-library-docs/ |
The first existing directory is used. If none exist, the primary location is used.
Environment variable override: Set MCP_LIBRARY_DOCS_CONFIG_DIR to use a custom location:
export MCP_LIBRARY_DOCS_CONFIG_DIR=/custom/path
Config File Format
Create config.yaml in the config directory to register external libraries:
# Global defaults (all optional)
defaults:
designs_dir: designs # Directory name to look for (default: designs)
index_file: INDEX.md # Index file name (default: INDEX.md)
cache: dynamic # "static" or "dynamic" (default: dynamic)
# External libraries
libraries:
# A shared utility library - Claude should import from this
lib-utils:
path: ~/projects/lib-utils
type: library # "library" = Claude can import from this
# Another project for reference - Claude should learn patterns, not import
other-app:
path: ~/projects/other-app
type: project # "project" = inspiration/patterns only
# Library with custom doc location
legacy-lib:
path: ~/projects/legacy-lib
designs_dir: docs/design # Override designs directory
index_file: README.md # Override index file name
cache: static # Cache on startup, don't re-read
Configuration Options
| Option | Default | Description |
|---|---|---|
path |
required | Path to library root (~ is expanded) |
type |
library |
library (import from) or project (patterns only) |
designs_dir |
designs |
Directory containing design docs |
index_file |
INDEX.md |
Name of the index file |
cache |
dynamic |
static (read once) or dynamic (read each time) |
Library Types
The type affects how Claude interprets the documentation:
library→[library]tag → "I can/should import and reuse this code"project→[project]tag → "Learn patterns, but don't import from here"- Current project →
[current project]tag → "I'm working here, prioritize this"
MCP Tools
list_libraries
Returns all INDEX.md contents from discovered and configured libraries.
Parameters:
cwd(string, required): Current working directory
Example response:
# my-app [current project]
> Main application
## Modules
### api
REST endpoints...
→ Full doc: api.md
---
# lib-utils [library]
> Shared utilities
## Modules
### dates
Date parsing utilities...
→ Full doc: dates.md
get_design_doc
Returns the full content of a specific design document.
Parameters:
library(string, required): Library name fromlist_librariesresponsetopic(string, required): Document name without.mdextension
Example: get_design_doc(library="lib-utils", topic="dates")
get_index_status
Returns the status of INDEX.md compared to actual files in designs/.
Parameters:
cwd(string, required): Current working directory
Returns: Shows which docs need adding to INDEX.md and which entries are stale.
Example: Use this before updating INDEX.md to see what needs attention.
MCP Prompts
update_index
Provides guidelines for updating INDEX.md files. Includes:
- Recommended format for entries
- Best practices (concise descriptions, key exports)
- Examples
Workflow for updating INDEX.md:
- Call
get_index_status(cwd)to see what needs updating - Get the
update_indexprompt for formatting guidance - Read any new .md files to understand them
- Update INDEX.md with properly formatted entries
Writing Good Design Docs
INDEX.md Structure
# {Library Name}
> One-line description
Install: `pip install {package}` or `import from {path}`
## Modules
### {module_name}
Brief description of what this module does.
Key exports: `export1`, `export2`, `export3`
→ Full doc: {module_name}.md
Topic Doc Structure
# {Module Name}
> One-line purpose
## When to use this
- Use case 1
- Use case 2
- **Don't use for**: [common mistakes]
## Key exports
### `function_name(param: Type) -> ReturnType`
What this function does.
\```python
result = function_name(value)
\```
## Patterns and conventions
[How to use this module correctly]
CLI Options
python -m mcp_library_docs [--debug] [--config PATH]
--debug: Enable debug logging to stderr--config PATH: Use custom config file path
Full Documentation
See designs/mcp-library-docs-design.md for complete design documentation.
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。