mcp-filesystem-extended
Enables bounded local filesystem inspection, comparison, and mutation with search and resume capabilities.
README
Filesystem MCP Server
TypeScript implementation of a local filesystem MCP server with bounded inspection, comparison, and mutation surfaces.
This root README is the DX-first entrypoint. It keeps only shared orientation and routes detailed tool guidance to endpoint-local README.md files.
Read this next
| Need | Start here |
|---|---|
| Shared conventions and cross-endpoint rules | CONVENTIONS.md |
| Shared architecture and ownership boundaries | DESCRIPTION.md |
| Tool-specific quick guidance | Endpoint-local README.md links below |
Architecture at a glance
| Layer | Responsibility |
|---|---|
application |
MCP bootstrap, tool-catalog composition, stable server framing, and server-scope exposure |
domain |
Tool-specific handlers, schemas, results, and runtime semantics |
infrastructure |
Path guarding, logging, persistence, and shared technical helpers |
Shared developer rules
- All path-based operations stay inside configured allowed directories.
- Broad-root discovery and recursive inspection default-exclude vendor, cache, and generated trees unless callers explicitly target them or reopen named descendants.
- Resume-capable inspection families stay same-endpoint and continue through
resumeTokenplus the appropriateresumeMode. - Primary result data stays complete in
content.text;structuredContentadds machine-readable envelope metadata and mirrored structured payloads. - Public read surfaces remain intentionally split:
read_files_with_line_numbersfor bounded inline batch reads,read_file_contentfor advanced single-file modes.
External dependency for content search
The regex, fixed-string, and native pattern-aware count lanes depend on the native ugrep executable.
This dependency is now resolved during MCP server startup preflight, not lazily during request execution. The server expects one of these runtime conditions:
UGREP_EXECUTABLE_PATHpoints to the native shell-freeugrepbinary, or- the MCP server process
PATHalready contains the directory that holds the nativeugrepexecutable.
A successful ugrep --version check in an interactive shell is helpful, but it is not sufficient by itself. The decisive environment is the Node.js process that runs the MCP server.
Common installation examples:
- Debian/Ubuntu:
apt-get install ugrep - Fedora/RHEL/CentOS:
dnf install ugrep - Arch:
pacman -S ugrep - macOS:
brew install ugrep - Windows:
winget install Genivia.ugreporchoco install ugrep
After installation, verify with ugrep --version and then restart the MCP client or IDE process that launches the server so the server runtime can inherit the updated environment.
Windows permanent fix when ugrep --version is not available
If ugrep is installed on Windows but ugrep --version still cannot be executed, you MUST apply the permanent Chocolatey fix below.
This usually means ChocolateyInstall is set incorrectly and Chocolatey created the package under the wrong root.
Run:
[Environment]::SetEnvironmentVariable('ChocolateyInstall','C:\ProgramData\chocolatey','Machine')
$env:ChocolateyInstall = 'C:\ProgramData\chocolatey'
$env:Path = "C:\ProgramData\chocolatey\bin;$env:Path"
choco install ugrep --force -y
Then open a new PowerShell window and verify again:
ugrep --version
If the MCP server still cannot start native search afterward, configure UGREP_EXECUTABLE_PATH with the absolute path to ugrep.exe for the process that launches the MCP server.
Endpoint README TOC
Application/server scope
Inspection — discovery
Inspection — metadata and integrity
Inspection — search family and count
inspection/searchfamily READMEsearch_file_contents_by_regexsearch_file_contents_by_fixed_stringcount_lines
Inspection — read
Comparison
Mutation — content
Mutation — path
Documentation boundary
Root documentation stays shared and non-redundant:
README.md= DX-first entrypointDESCRIPTION.md= architecture indexCONVENTIONS.md= shared conventions and leaf-slice routing- endpoint-local
README.mdfiles = detailed developer-facing guidance per public endpoint
This root file is intentionally a navigation surface, not a second endpoint-by-endpoint manual.
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。