That is the feature that separates this from a one-shot bridge: a single JSON config in the `mcpServers` shape you already have gives each server its own `/servers/NAME/` path, plus a `/status` endpoint to check them. Note the config file wins outright — when `--named-server-config` is supplied, the individual `--named-server` flags are ignored.
A Python transport proxy with two modes. One turns a remote SSE or Streamable HTTP server into a local stdio process, which is how a client like Claude Desktop reaches a server it cannot speak to natively. The other opens a port, spawns local stdio servers, and serves them over SSE.
- Stdio to SSE or Streamable HTTP, choosing the protocol with `--transport` set to `sse` or `streamablehttp`
- Headers for authenticated remotes via `--headers`, or the `API_ACCESS_TOKEN` variable as a shorthand for a bearer token
- OAuth2 client credentials support through `--client-id`, `--client-secret` and `--token-url`
- SSE to stdio the other way, with `--sse-port` and `--sse-host` controlling where it listens and `--env` passing variables to the spawned server
- Named servers: several local servers behind one port, each at `/servers/NAME/`, defined by repeated `--named-server` flags or a single `--named-server-config` JSON file in the familiar `mcpServers` shape
- A default server at the root paths alongside the named ones, and a global `/status` endpoint
Python, installed from PyPI with uv or pip, from the GitHub repository for the latest, or as a container. The client launches `mcp-proxy` with the remote URL as the first argument; in the other direction the command for the local server goes after a `--` separator. Version 0.12.0.
One command — uv tool install mcp-proxy