mcpproxy-go
Supercharge AI Agents, Safely
claude mcp add --transport stdio smart-mcp-proxy-mcpproxy-go mcpproxy serve
How to use
MCPProxy is a desktop gateway that aggregates multiple MCP servers and serves as a single point of interaction for AI agents. It discovers and proxy-distributes tools from connected MCP servers, helping you bypass per-server tool limits and manage the set of available tools securely. Once running, you can connect your AI tooling or IDEs to the local proxy (default http://127.0.0.1:8080) and perform tool discovery, retrieve_tools calls, and tool execution through a unified interface. The proxy also provides a system tray UI for server management, health checks, and an option to quarantine untrusted servers before enabling them in your workflow. Tools and servers are configured in mcp_config.json, where you specify how each upstream MCP server is reached (local processes, HTTP endpoints, etc.).
How to install
Prerequisites:
- A supported operating system (Windows, macOS, Linux)
- Optional: Go toolchain or a prebuilt MCPProxy binary for your platform
Install options:
- Install via binary release (recommended for most users)
- Go to the MCPProxy releases page and download the latest binary equivalent for your OS (e.g., mcpproxy).
- Ensure the binary is executable and accessible in your PATH.
- Install via Homebrew (macOS)
- brew install smart-mcp-proxy/mcpproxy/mcpproxy
- Install from source (Go required)
- Ensure Go 1.20+ is installed
- Run: go install github.com/smart-mcp-proxy/mcpproxy-go/cmd/mcpproxy@latest
- Ensure $GOPATH/bin or $HOME/go/bin is in your PATH
- Run the proxy
- After installation, start the proxy: mcpproxy serve
- The proxy will listen on http://127.0.0.1:8080 by default and show a system tray UI for management
- Configure servers
- Create or edit the configuration file at ~/.mcpproxy/mcp_config.json to add upstream MCP servers (see the example in the repository's docs) and then reload or restart the proxy as needed.
Additional notes
Tips and common considerations:
- Security: Enable TLS and client certs if exposing the proxy beyond localhost. Use the tls settings in mcp_config.json and consider using mcpproxy trust-cert to install the CA locally.
- Quarantine: Untrusted or unknown MCP servers can be quarantined until manually approved. This helps prevent Tool Poisoning Attacks.
- Performance: Adjust top_k, tools_limit, and tool_response_limit to balance discovery speed with resource use.
- Upstream management: Use the CLI under mcpproxy upstream to restart, enable, disable, or view logs for individual servers, or apply actions across all servers with --all.
- Debugging: Use mcpproxy tools list --server=NAME to test tool discovery for a specific server before going into production.
- When integrating with IDEs or external tools, point them to the local proxy endpoint (e.g., http://localhost:8080/mcp/) and rely on the proxy to mediate tool calls.
Related MCP Servers
aci
ACI.dev is the open source tool-calling platform that hooks up 600+ tools into any agentic IDE or custom AI agent through direct function calling or a unified MCP server. The birthplace of VibeOps.
automagik-genie
🧞 Automagik Genie – bootstrap, update, and roll back AI agent workspaces with a single CLI + MCP toolkit.
go-utcp
Official Go implementation of the UTCP
CodeMCP
Code intelligence for AI assistants - MCP server, CLI, and HTTP API with symbol navigation, impact analysis, and architecture mapping
floop
Spreading activation memory for AI coding agents - corrections in, context-aware behaviors out.
slack
Session-based Slack MCP for Claude and MCP clients: local-first workflows, secure-default HTTP.