Overview
Configuration workflow for setting up a single MCP gateway daemon that fronts all upstream MCP servers. Each AI assistant points at the daemon instead of spawning its own MCP fleet.
Trigger
One-time setup, plus rotation every config change
Schedule
One-time setup
Workflow Steps
1
Install MCP gateway
FastMCP, @modelcontextprotocol/server-gateway, or equivalent.
2
List upstream MCP servers
Scavio MCP, Postgres MCP, GitHub MCP, internal MCPs.
3
Define gateway config
JSON listing each upstream server's URL or command.
4
Run gateway as HTTP daemon
One process serving all agents on a known port.
5
Point agents at the daemon
Claude Desktop, Cursor, opencode all reference http://localhost:8765/mcp.
6
Verify consolidation
ps aux | grep mcp should show 1, not 30+.
Python Implementation
Python
# Configuration-driven setup; no Python code needed.
# See gateway.json and per-agent mcp config below.JavaScript Implementation
JavaScript
// gateway.json
{
"upstreams": {
"scavio": { "url": "https://mcp.scavio.dev/mcp", "headers": { "x-api-key": "$SCAVIO_API_KEY" } },
"postgres": { "command": "npx", "args": ["@modelcontextprotocol/server-postgres", "$DATABASE_URL"] }
}
}
// Then in claude_desktop_config.json / .cursor/mcp.json / opencode config:
{
"mcpServers": {
"gateway": { "url": "http://localhost:8765/mcp" }
}
}Platforms Used
Web search with knowledge graph, PAA, and AI overviews