mcp-remote-py
A minimal Python-based proxy that bridges local MCP STDIO clients with remote MCP SSE servers. It enables bidirectional JSON-RPC message passing between standard command-line tools and web-based remote endpoints.
README
mcp-remote-py
Minimal MCP STDIO ↔ Remote SSE proxy in Python.
This project is a deliberately small reimplementation of the core idea from mcp-remote:
- Local side: MCP over STDIO (newline-delimited JSON / JSONL)
- Remote side: MCP over SSE (receives JSON-RPC via SSE
messageevents) - Send path: remote provides an
endpointSSE event; subsequent client→server messages are sent via HTTP POST to that endpoint.
What this is / isn’t
Included
- Bidirectional proxying of MCP JSON-RPC messages
- No stdout contamination (logs go to stderr)
- Optional custom headers (
--header 'Name: value')
Intentionally excluded (keep it simple)
- OAuth / authorization flows
- Token storage / refresh
- Multi-instance coordination / lockfiles
- HTTP↔SSE fallback strategies
Install
Option A — “Global” install with uv (recommended)
Installs a user-scoped executable into ~/.local/bin (macOS/Linux) or %USERPROFILE%\.local\bin (Windows).
chmod +x ./*.sh
./install-global-uv.sh
Windows (CMD):
install-global-uv.cmd
Uninstall:
./uninstall-global-uv.sh
Windows (CMD):
uninstall-global-uv.cmd
If your shell can’t find the command, ensure ~/.local/bin is on your PATH.
Option B — Dev install (editable)
python -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -e .
Run
mcp-remote-py http://127.0.0.1:8080/mcp/sse
Transport strategy (Node mcp-remote와 유사):
# default: http-first
mcp-remote-py http://127.0.0.1:8080/mcp/sse --transport http-first
# force SSE-only (서버가 HTTP transport를 지원하지 않을 때)
mcp-remote-py http://127.0.0.1:8080/mcp/sse --transport sse-only
Custom headers:
mcp-remote-py http://127.0.0.1:8080/mcp/sse \
--header 'Authorization: Bearer TOKEN'
MCP client configs (copy/paste)
Codex (~/.codex/config.toml)
Codex expects TOML like this:
[mcp_servers.ida-remote]
command = "/Users/<you>/.local/bin/mcp-remote-py"
args = ["http://127.0.0.1:8080/mcp/sse"]
If you installed via uv into ~/.local/bin, command is typically:
command = "/Users/<you>/.local/bin/mcp-remote-py"
Claude Desktop (claude_desktop_config.json)
{
"mcpServers": {
"ida-remote": {
"command": "mcp-remote-py",
"args": ["http://127.0.0.1:8080/mcp/sse"]
}
}
}
Cursor (~/.cursor/mcp.json)
{
"mcpServers": {
"ida-remote": {
"command": "mcp-remote-py",
"args": ["http://127.0.0.1:8080/mcp/sse"]
}
}
}
Windsurf (~/.codeium/windsurf/mcp_config.json)
{
"mcpServers": {
"ida-remote": {
"command": "mcp-remote-py",
"args": ["http://127.0.0.1:8080/mcp/sse"]
}
}
}
Troubleshooting
- Nothing works / random parse errors: ensure the remote SSE server emits JSON-RPC only on the default
messageevent. Non-protocol events are ignored. - Don’t see the command: add
~/.local/bintoPATH(macOS zsh:~/.zprofileis a good default). - Logs: all logs go to stderr by design; stdout is reserved for MCP protocol.
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。