Multi-Server Aggregation
Aggregate any number of external MCP servers behind a single unified endpoint. Manage lifecycle, restart on crash, and route calls transparently.
A unified MCP gateway that brings order to your tool chaos.
Managing multiple MCP servers means juggling separate processes, colliding tool names, and no visibility into what's healthy and what's failing. Porter unifies them behind a single gateway with automatic namespacing and built-in health monitoring — so you can focus on what your tools do, not how they connect.
One gateway. Every MCP server. Zero collisions.
Aggregate any number of external MCP servers behind a single unified endpoint. Manage lifecycle, restart on crash, and route calls transparently.
Collision-free tool names via automatic slug:tool_name prefixing. No more silent overwrites when two servers expose the same tool.
Three-tier health model — healthy, degraded, unhealthy. Graceful degradation removes failing servers from the tool surface without crashing the gateway.
Connect via STDIO subprocesses, HTTP/SSE for remote servers, or Streamable HTTP for streaming responses. One gateway, every transport.
Edit your config and Porter picks up changes automatically — no restart required. Connected MCP clients are notified of the updated tool surface instantly.
Environment variables use ${VAR} reference syntax — secrets never appear in config files. Validation catches misconfigurations before any servers are spawned.
MCP servers get access to your tools, your tokens, and your AI assistant's context. Porter treats that seriously.
Environment variables must use $${VAR} reference syntax. Porter validates this at startup and rejects literal values — your tokens never end up committed to a repo.
Config is validated before any server is spawned. Duplicate slugs, missing required fields, mixed transport types, and bare secrets are all caught immediately — not after half your servers are already running.
Every tool is prefixed with its server's slug. An AI assistant can't accidentally call the wrong server's tool because names that look similar in different servers are always distinct in Porter's namespace.
Servers that exceed a 50% error rate are automatically excluded from tool listings and blocked from receiving calls. Your AI assistant never sees tools from a server that can't reliably handle them.