spm-search-mcp
An MCP server that allows coding agents to search the Swift Package Index and retrieve GitHub READMEs without requiring an API key. It features comprehensive search filters for stars, platforms, and licenses, providing token-efficient responses optimized for LLM comprehension.
README
spm-search-mcp
An MCP server that lets coding agents search the Swift Package Index. No API key required.
Built with FastMCP and designed using arcade patterns for optimal agent comprehension.
Quick install
The fastest way to add this server to your MCP client:
# Claude Desktop
uvx fastmcp install claude-desktop spm_search_mcp/server.py:mcp -n "Swift Package Index"
# Claude Code
uvx fastmcp install claude-code spm_search_mcp/server.py:mcp -n "Swift Package Index"
# Cursor
uvx fastmcp install cursor spm_search_mcp/server.py:mcp -n "Swift Package Index"
# Any client — get the JSON snippet to paste
uvx fastmcp install mcp-json spm_search_mcp/server.py:mcp -n "Swift Package Index"
Manual MCP client configuration
Published package (once on PyPI)
{
"mcpServers": {
"spm-search": {
"command": "uvx",
"args": ["spm-search-mcp"]
}
}
}
From source (development)
{
"mcpServers": {
"spm-search": {
"command": "uv",
"args": [
"run",
"--with", "fastmcp",
"--with-editable", "/path/to/spm-search",
"fastmcp", "run",
"/path/to/spm-search/src/spm_search_mcp/server.py:mcp"
]
}
}
}
Tools
search_swift_packages
Search for Swift packages by keyword, author, stars, platform, license, and more. All of SPI's filter syntax is exposed as typed parameters — agents never need to learn the DSL.
| Parameter | Type | Description |
|---|---|---|
query |
str |
Free-text search (e.g. "networking", "json parsing") |
author |
str |
Repository owner (e.g. "apple", "vapor") |
keyword |
str |
Package keyword tag (e.g. "server", "ui") |
min_stars |
int |
Minimum GitHub star count |
max_stars |
int |
Maximum GitHub star count |
platforms |
list |
Compatible platforms: ios, macos, watchos, tvos, visionos, linux |
license_filter |
str |
"compatible" for App Store, or SPDX ID like "mit", "apache-2.0" |
last_activity_after |
str |
ISO8601 date — only active packages (e.g. "2024-01-01") |
last_commit_after |
str |
ISO8601 date — only recently committed packages |
product_type |
str |
library, executable, plugin, or macro |
page |
int |
Page number for pagination (default 1) |
All parameters are optional. At least one must be provided. Parameters combine with AND logic.
get_package_readme
Fetch a package's README from GitHub. Returns truncated content by default (4000 chars) to save tokens.
| Parameter | Type | Description |
|---|---|---|
owner |
str |
GitHub repository owner (e.g. "Alamofire") |
repo |
str |
GitHub repository name (e.g. "Alamofire") |
max_length |
int |
Max chars to return (default 4000). Set to 0 for full content. |
Example usage
Once connected, an agent can:
# Search for networking libraries with 500+ stars
search_swift_packages(query="networking", min_stars=500)
# Find iOS-compatible packages
search_swift_packages(platforms=["ios"], min_stars=100)
# Browse a specific author's packages
search_swift_packages(author="apple")
# Read a package's README
get_package_readme(owner="Alamofire", repo="Alamofire")
# Get full README (no truncation)
get_package_readme(owner="apple", repo="swift-nio", max_length=0)
Error handling
All errors return structured, actionable messages instead of raw exceptions:
- RETRYABLE — transient failures (timeouts, rate limits, server errors). The agent can retry.
- PERMANENT — the request itself is wrong (404, 403). The agent should change its approach.
Every error message tells the agent what happened, why, and how to fix it.
Design
This server implements these arcade patterns:
- QUERY_TOOL — both tools are read-only, safe to retry
- CONSTRAINED_INPUT — enums for platforms and product types
- SMART_DEFAULTS — all parameters optional with sensible defaults
- NEXT_ACTION_HINT — every response suggests what to do next
- GUI_URL — every result includes SPI + GitHub URLs
- TOKEN_EFFICIENT_RESPONSE — truncated README with opt-in full
- PAGINATED_RESULT —
has_moreflag with page navigation - ERROR_CLASSIFICATION — RETRYABLE vs PERMANENT error tagging
- RECOVERY_GUIDE — actionable error messages with fix instructions
- PROGRESSIVE_DETAIL —
max_length=0for full README content
Development
uv sync # install dependencies
uv run pytest # run tests (62 tests)
uv run poe test-cov # run with coverage (90%+ required)
uv run ruff check . # lint
uv run ty check . # type check
prek run --all-files # run all pre-commit hooks
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。