The namespacing is the whole design: `accuweather___getWeather` is ugly to read and unambiguous to call, which is the right trade when a dozen upstream servers are sharing one connection. Use server mode when you know exactly which one server you want and would rather see its tools under their real names. The catch with any aggregator is that the tool list you see belongs to the servers behind it, not to this package — what this package itself contributes is the four discovery tools and the routing.
The client-side proxy for MCP-Hive, a hosted catalogue of MCP servers. It runs in your agent host and routes calls out to Hive, in one of two modes: server mode proxies a single named MCP server through to your client, and gateway mode exposes every server available to you through one connection, with each upstream tool namespaced as `serverName___toolName`.
- A single named MCP server from Hive wired into your client, with `--server <name>` naming which one
- Or every server at once through one connection with `--gateway`, where tools arrive namespaced — `validator___echo`, `accuweather___getWeather` — so two servers that both call a tool `search` do not collide
- Discovery tools in gateway mode that let the agent find out what is reachable before it calls anything: `discoverServers`, `listTools`, `listResources`, `listPrompts`
- A choice of transport: stdio by default, or HTTP with `--http` and `--port`, plus `--verbose` when a call is not going where you expect
Consumer credentials from mcp-hive.com, passed with `--credentials`. The npm package is `@mcp-hive-utils/proxy` (1.0.19 in package.json, 1.0.15 pinned in server.json) with a `mcp-hive-proxy` binary and stdio transport; a Smithery listing is available for one-click install into Claude Desktop. Whatever the upstream server itself needs — its own API key, its own account — still applies; the proxy carries your Hive identity, not the upstream service's.
One command plus a key — npx -y @mcp-hive-utils/proxy, then supply credentials