Click on "Deploy 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., "@proxy-mcp find a tool to send a slack message"
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.
Universal MCP proxy server that discovers, searches, and executes tools across all your configured MCP servers from a single entry point.
Proxy execution — Call any tool on any MCP server through execute_mcp_tool
Semantic search — Find tools by describing what you want to do, not by exact name
Connection pool — MCP subprocesses are spawned on-demand and reused across calls
Weight learning — The search system learns from your selections to improve rankings over time
Auto-indexing — All MCP tools are indexed on server startup (non-blocking)
Related MCP server: mcpproxy-go
Tool | Description |
| Execute any tool on any configured MCP server |
| Semantic search across all indexed tools |
| List configured MCP servers and their status |
| Re-index tools from MCP servers |
| View learned search weights |
| Reset all learned weights |
| Close all active MCP connections |
Python >= 3.10
uv (recommended) or pip
# Clone
git clone https://github.com/Rixmerz/proxy-mcp.git
cd proxy-mcp
# Install with uv
uv syncAdd to ~/.claude.json under mcpServers:
{
"mcpServers": {
"proxy-mcp": {
"command": "uv",
"args": ["run", "--directory", "/path/to/proxy-mcp", "proxy-mcp"]
}
}
}The proxy reads MCP server configs from:
~/.agentcockpit/mcps.json (priority)
~/.claude.json (fallback)
# Search for a tool by what you want to do
search_tools(query="expose service to internet")
# Execute a tool on a specific MCP
execute_mcp_tool(
mcp_name="Context7",
tool_name="get-library-docs",
arguments={"context7CompatibleLibraryID": "/vercel/next.js", "topic": "routing"}
)
# List all available MCPs
list_available_mcps()Learned weights are stored at ~/.proxy-mcp/learned_weights.json.
MIT
Close all active MCP connections.
Use this to clean up resources when done with MCP tools. Connections will be re-established on next use.
Inputs: no named parameters. Returns: object
Execute any available MCP tool through the proxy.
This is the universal gateway for calling MCP tools. The proxy spawns MCP servers on-demand and maintains a connection pool for efficient reuse. MCP configurations are read from ~/.claude.json or ~/.agentcockpit/mcps.json.
Inputs: mcp_name (required), arguments (required), tool_name (required). Returns: object
View the learned weights from the search system.
Shows which tools have been selected and for which keywords, revealing how the system has learned from your selections.
Inputs: top_n, tool_filter. Returns: object
List all configured MCP servers and their connection status.
Returns the names of all MCP servers found in configuration, along with which ones have active connections.
Inputs: no named parameters. Returns: object
Refresh the tool index for semantic search.
Connects to MCP servers and retrieves their tool lists for indexing. Use after adding new MCPs. The index is auto-loaded on server startup.
Inputs: mcp_name. Returns: object
Reset all learned weights.
WARNING: This erases all accumulated learning.
Inputs: confirm. Returns: object
Search tools by objective or description using semantic similarity.
Useful when you don't know the exact tool name but know what you want to do.
Inputs: query (required), mcp_filter, max_results. Returns: object
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 7 tools
Each tool has a clearly distinct purpose: execution, search, server listing, weight management, index refresh, and connection cleanup. There is no functional overlap between any pair of tools.
All tool names follow a uniform verb-first snake_case pattern (e.g., execute_mcp_tool, list_available_mcps, reset_learned_weights). The verbs are action-oriented and consistently describe the operation.
Seven tools is well-scoped for a proxy MCP server covering discovery, execution, learning, and maintenance. Each tool earns its place without redundancy or bloat.
The tool set covers all essential operations for the proxy's purpose: finding tools (search_tools, refresh_tool_index), executing tools (execute_mcp_tool), managing servers (list_available_mcps, close_mcp_connections), and controlling learning (get/reset_learned_weights). No significant gaps are evident.
MCP server for progressive tool usage at any scale (see https://klavis.ai)
The MCP server that finds MCP servers. Aggregates Official Registry, Glama, and Smithery.
Single-connection MCP gateway discovering and executing 490+ verified MCP & A2A tools.
MCP federation gateway: route, govern and observe agent tools across clouds and MCP servers.
