biome-mcp
Enables LLMs to run Biome formatting, linting, and assist checks on TypeScript/web projects with token-efficient, structured output, plus searchable cached Biome documentation.
README
biome-mcp
A Model Context Protocol (MCP) server for running Biome on TypeScript and web projects with token-efficient output.
Overview
biome-mcp gives LLMs direct access to Biome's formatting, linting, and assist checks without shelling out to the CLI. Diagnostic output is parsed structurally and reduced to compact, grouped results before it reaches the model, so long check runs stay readable and cheap.
The server also caches the Biome CLI, configuration, and JavaScript linter documentation and exposes it as searchable resources.
Features
- biome_check - Run formatting, linting, and assist checks in one pass
- biome_lint - Lint-only diagnostics with an
all/error/warningseverity filter - biome_format - Verify formatting without modifying files
- Structured parsing -
json-prettyoutput avoids fragile parsing of human-readable text - Adaptive aggregation - large result sets include rule/file rollups before flat diagnostics
- changed_only - Check only staged, unstaged, and untracked web files (
*.ts,*.tsx,*.js, ...) - Documentation cache - Searchable, version-aware cache of Biome docs
- Bundled Agent Skill - Install
biome-mcpskill into.agents/skillswith one command
How It Works
biome-mcp implements the Model Context Protocol
to expose Biome through read-only tools. Every check tool returns a
CheckResult with compact output text and the underlying exit_code; it
never writes to source files.
| Tool | What it runs |
|---|---|
biome_check(paths?, changed_only?) |
biome check (format + lint + assist) |
biome_lint(paths?, level?, changed_only?) |
biome lint, filtered by severity |
biome_format(paths?, changed_only?) |
biome check with linter and assist disabled |
search_documentation(query) |
Search the cached Biome docs |
update_cache(force?) |
Refresh the cached docs from biomejs.dev |
Usage
To start the server:
uvx biome-mcp
Or from a checkout:
uv run biome-mcp
Configure an MCP client to launch the server in the project directory to check, e.g. for opencode:
{
"mcp": {
"biome-mcp": {
"type": "stdio",
"command": "uvx",
"args": ["biome-mcp"]
}
}
}
The server checks the project in its working directory. To target another
directory, set BIOME_MCP_PROJECT_DIR.
Biome is invoked through bunx @biomejs/biome by default. Set
BIOME_MCP_COMMAND_PREFIX to use another project-local invocation (see
docs/configuration.md).
Agent Skill
Install the bundled biome skill in a project when the agent client discovers
skills from .agents/skills:
uvx biome-mcp skill install biome --target .
The skill directs agents to the server's read-only tools and the cached documentation resources.
Configuration
Environment variables, all optional:
| Variable | Default | Purpose |
|---|---|---|
BIOME_MCP_LOG_LEVEL |
INFO |
Logging level (stderr) |
BIOME_MCP_PROJECT_DIR |
server working directory | Project directory to check |
BIOME_MCP_COMMAND_PREFIX |
bunx @biomejs/biome |
Invocation for Biome; empty string uses binaries from PATH |
BIOME_MCP_COMMAND_TIMEOUT |
none | Optional per-command timeout in seconds |
BIOME_MCP_CACHE_DIR |
~/.cache/biome-mcp/docs |
Documentation cache location |
Documentation
See the docs for the full picture:
- Architecture - components, request flow, and output processing
- Configuration - environment variables and client setup
- Tools & resources - complete catalog with signatures
- Development - layout, tests, and contributing
Development
This project is built with FastMCP and uv.
uv sync --extra dev
uv run ruff check
uv run ruff format --check
uv run ty check
uv run pytest
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。