mcporter-bridge
Allows reading and inspecting Figma design files through the Figma MCP server.
Enables management of GitHub repositories, including file operations, issues, PRs, and code search.
Provides tools to read and write Notion pages.
Facilitates processing and conversion of WeChat articles.
Enables searching and publishing notes on Xiaohongshu (Little Red Book).
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@mcporter-bridgelist all available MCP servers"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
mcporter-bridge
Expose your local MCPorter registry as one stable MCP server for coding clients.
mcporter-bridge is a small FastMCP server that turns your existing mcporter setup into a single, reusable entry point for clients like Codex, Claude Code, Cline, and Cursor.
Why
If you use multiple coding clients, your MCP setup usually fragments fast:
Codex has one config format
Claude Code has another
Cline and Cursor add their own
mcporteralready knows your real server registry, auth state, and runtime
mcporter-bridge keeps mcporter as the source of truth and gives clients one stable MCP server instead of another pile of duplicated configs.
What It Does
Reads your local
mcporterregistry at runtimeLists configured MCP servers and their health
Inspects a specific server and its tool schemas
Calls any tool on any configured server through
mcporterLazy loading: Activate/deactivate heavy MCPs on demand to save context
Optionally exposes local
agent-reachdiagnostics as helper tools
This is not a transport proxy and not an enterprise gateway. It is a local-first bridge for client integration.
Tools
Core Tools
Tool | Description |
| 【MCP 全景】列出所有已加载(active)和可激活(available)的 MCP 服务器,包含类型区分(small/heavy) |
| 探索 MCP 服务器的工具列表和参数格式,支持 |
| 调用指定 MCP 的某个工具 |
Lazy Loading Tools (按需加载)
Tool | Description |
| 激活一个大型 MCP(如 chrome-devtools, playwright) |
| 停用大型 MCP 释放上下文 |
Utility Tools
Tool | Description |
| 检查 mcporter 状态(config doctor / version) |
All tools return structured output with:
oktimed_outcommandtimeout_msreturncodestdoutstderrparsed_json
Install
pip
pip install mcporter-bridgepipx
pipx install mcporter-bridgeAfter installation, you get two commands:
mcporter-bridgeto run the MCP servermcporter-bridge-configto generate or install client snippets
local development
git clone https://github.com/Citrus086/mcporter-bridge.git
cd mcporter-bridge
python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"Requirements
mcporterinstalled and available onPATHat least one configured MCP server in your local
mcporterregistry
Optional:
Quick Check
mcporter-bridgeIn another shell:
mcporter list --stdio "python3 -m mcporter_bridge" --jsonClient Setup
Ready-made templates live in examples/.
Codex
Add this to ~/.codex/config.toml:
[mcp_servers.mcporter-bridge]
type = "stdio"
command = "python3"
args = ["-m", "mcporter_bridge"]
startup_timeout_ms = 30000Or install it automatically:
mcporter-bridge-config install --client codexNotes:
OpenAI’s docs confirm that Codex reads MCP config from
~/.codex/config.tomlusing themcp_servers.<name>structure for MCP servers.The stdio form here is also verified locally with
codex mcp add --help, which acceptscodex mcp add <name> -- <command...>for stdio servers.
Claude Code / Claude Desktop
Add this to your MCP config:
{
"mcpServers": {
"mcporter-bridge": {
"command": "python3",
"args": ["-m", "mcporter_bridge"]
}
}
}Automatic install using the default ~/.claude.json path:
mcporter-bridge-config install --client claudeNotes:
Anthropic’s Claude Code docs confirm that user-scoped MCP servers live in
~/.claude.json.The same docs show project-scoped servers stored in
.mcp.jsonat the project root.The documented JSON shape uses
mcpServerswithcommand,args, andenvfor stdio servers.
Cline
Use the same stdio shape:
{
"mcpServers": {
"mcporter-bridge": {
"command": "python3",
"args": ["-m", "mcporter_bridge"]
}
}
}Generate snippets:
mcporter-bridge-config snippet --client clineWrite to an explicit config path:
mcporter-bridge-config install --client cline --config-path /path/to/mcp.jsonNotes:
Cline’s docs confirm that MCP settings are stored in
cline_mcp_settings.json.For local stdio servers, the documented JSON shape uses
mcpServerswithcommand,args,env,alwaysAllow, anddisabled.Cline’s docs use
typefor remote transport config such asstreamableHttp, but not in the local stdio example, so the bridge does not emittypefor Cline.
Cursor
Cursor uses mcp.json with mcpServers, and for stdio servers the bridge emits type: "stdio" explicitly.
Global config example:
{
"mcpServers": {
"mcporter-bridge": {
"type": "stdio",
"command": "python3",
"args": ["-m", "mcporter_bridge"]
}
}
}Install into the default global path:
mcporter-bridge-config install --client cursorOr write to an explicit config path:
mcporter-bridge-config install --client cursor --config-path /path/to/mcp.jsonConfig Helper
Print a snippet:
mcporter-bridge-config snippet --client codexCustomize the launcher:
mcporter-bridge-config snippet \
--client claude \
--python-command /opt/homebrew/bin/python3.13 \
--module-name mcporter_bridgeInstall directly into a config file:
mcporter-bridge-config install --client codex
mcporter-bridge-config install --client claudeWhen an existing config file is updated, the helper writes a sibling backup with a .bak suffix first.
Example Prompts
"列出所有可用的 MCP 服务器"
"查看 xiaohongshu 服务器的工具"
"调用 xiaohongshu 的 check_login_status"
"激活 playwright 浏览器工具"
"查看 xiaohongshu 的原始工具定义(raw=true)"
Lazy Loading (按需加载)
大型 MCP 如 chrome-devtools (29 tools) 或 playwright (22 tools) 会消耗大量上下文。默认保持未加载状态,需要时再激活。
MCP 类型
small - 小型 MCP,通常已加载,可直接使用
heavy - 大型 MCP,默认未加载,需先激活
Setup
创建 heavy MCP 目录:
mkdir -p ~/.mcporter/heavy/available将大型 MCP 从
mcporter.json移到单独文件:
# 示例:将 playwright 移到 heavy
echo '{
"mcpServers": {
"playwright": {
"command": "npx",
"args": ["-y", "@playwright/mcp@latest"],
"description": "浏览器自动化",
"tags": ["浏览器"]
}
}
}' > ~/.mcporter/heavy/available/playwright.json从主
mcporter.json中移除该 MCP。
Workflow
1. mcporter_list_servers()
↓ 返回 active(已加载)和 available(可激活)两个列表
↓ 需要的 MCP 在 available 中?
2. mcporter_activate_mcp(name="playwright")
↓
3. mcporter_list_servers() # 确认已出现在 active 列表
↓
4. 使用 playwright 工具...
↓
5. mcporter_deactivate_mcp(name="playwright") # 用完即释放这让上游 LLM 能够自我管理上下文,按需加载/卸载大型 MCP。
Server Descriptions (服务器描述)
mcporter-bridge 会为每个 MCP 服务器提供功能描述,帮助 LLM 了解每个服务是干什么的。
描述来源(优先级从高到低)
用户配置 - 在
~/.mcporter/mcporter.json中添加内置映射 - 常见 MCP 的预设描述
名字推断 - 从服务器名称猜测功能
在 mcporter.json 中配置描述
{
"mcpServers": {
"my-custom-mcp": {
"command": "...",
"args": [...],
"description": "自定义 MCP 的功能描述",
"tags": ["标签1", "标签2"],
"best_for": "最适合的使用场景"
}
}
}内置描述的常见 MCP
服务器名 | 描述 | 用途 |
exa | AI 搜索引擎 | 搜索网页内容、代码、新闻 |
xiaohongshu | 小红书操作 | 搜索/发布小红书笔记 |
douyin | 抖音操作 | 解析抖音视频、下载无水印视频 |
web-search-prime | 智谱网页搜索 | 搜索中文网页内容 |
web-reader | 网页阅读器 | 读取并解析网页内容 |
context7 | 技术文档查询 | 查询编程库的官方文档 |
zread | GitHub 仓库阅读 | 搜索/阅读 GitHub 仓库内容 |
zai-mcp-server | Z.AI 多模态工具 | 图像分析、视频分析、OCR |
notion | Notion 笔记操作 | 读写 Notion 页面 |
figma | Figma 设计工具 | 读取 Figma 设计稿 |
github | GitHub 官方 MCP | 创建/更新文件、管理 Issues/PRs、搜索代码 |
linkedin-scraper | LinkedIn 数据抓取 | 获取个人/公司资料、搜索职位和人才 |
wechat-article | 微信文章处理 | 转换微信文章格式 |
名字推断规则
如果服务器不在内置列表中,会从名称推断:
包含
search→ 搜索相关包含
browser/chrome→ 浏览器相关包含
github/git→ 代码仓库相关包含
xiaohongshu→ 小红书相关...
Environment Variables
MCPORTER_BRIDGE_MCPORTER_BIN: override themcporterbinary pathMCPORTER_BRIDGE_MAX_OUTPUT_CHARS: cap captured stdout/stderr length
Troubleshooting / 一个跨三层的隐蔽 bug
如果你在 macOS + Anaconda Python 环境下遇到 ModuleNotFoundError: No module named mcporter_bridge,或者 Claude 一直报 Failed to reconnect,问题可能来自一个跨三层的隐蔽 bug:
pip 写入 .pth 文件
↓
macOS 自动给该文件打上 UF_HIDDEN flag(因为父目录 .venv 被标记为隐藏)
↓
Anaconda Python 的 site.py 读取 .pth 时,检测到 UF_HIDDEN 就直接跳过
↓
Editable install 失效,import mcporter_bridge 失败这会导致 python3 -m mcporter_bridge 直接起不来进程,Claude 等 MCP 客户端自然无法重连。
我们已经做的修复
__main__.py里加了 bootstrap fallback:如果模块加载失败,会自动探测src/目录并注入sys.path。mcporter-bridge-config在生成客户端配置时,自动检测 editable install 并写入PYTHONPATH,同时写入FASTMCP_SHOW_SERVER_BANNER=false保证 stdio 干净。show_banner=False:确保 FastMCP 的启动 banner 不会污染 stdio 的 JSON-RPC 输出。
临时 workaround
如果你不想重新安装,可以手动去掉 hidden flag:
chflags nohidden /path/to/your/.venv/lib/python3.12/site-packages/_mcporter_bridge.pth或者直接给 Claude/Codex 等客户端的 env 加上:
"env": {
"PYTHONPATH": "/absolute/path/to/mcporter-bridge/src",
"FASTMCP_SHOW_SERVER_BANNER": "false"
}这个 bug 的诡异之处在于:pip、macOS、Python 三层单独看都没错,但串在一起就把 editable install 干掉了。更让人意外的是,我们项目的 build backend
hatchling目前并不支持 symlink editable mode,所以即使想绕过.pth文件也做不到——它只能生成.pth文件。目前 Python 和 macOS 官方 issue tracker 里似乎还没有人专门报告这一现象。
Roadmap
higher-level convenience tools for common MCPs
optional tool allowlists / denylists
more client-specific path discovery
richer auth and diagnostics helpers
License
MIT
This server cannot be installed
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/Citrus086/mcporter-bridge'
If you have feedback or need assistance with the MCP directory API, please join our Discord server