Summary

Validate before you launch — it is the command that saves the wasted startup.

Most orchestration failures are mundane: a command missing from PATH, or an environment variable that was never set. The validate step reports both before anything starts, and separates fatal errors from warnings so a missing optional variable does not read like a broken config. Environment variable lookup is case-insensitive, so a config naming github_token also matches GITHUB_TOKEN.

What it is

A command-line orchestrator. Point it at a config file listing your MCP servers and it installs each one, launches them together, and fronts them all with one HTTP gateway — so clients that speak HTTP reach every tool through a single address instead of spawning processes.

What you get

`fluidmcp run config.json --file --start-server` installs and launches everything in the config behind one gateway at http://localhost:8099
Automatic API documentation at /docs, so the whole tool surface is browsable
Per-server HTTP endpoints that accept JSON-RPC tools/call directly, plus SSE streaming with sse/start, sse/stream, sse/message and sse/tools_call
Package management: `fluidmcp install author/package@version`, `fluidmcp list`, and `fluidmcp edit-env` to change a package's environment
`fluidmcp validate` checks config structure, whether each command exists on PATH, and whether required environment variables are set — distinguishing fatal errors from warnings by exit code
Bearer token authentication with `--secure --token`, and configs loadable from an S3 URL

Requirements

Python — `pip install fluidmcp`. A `config.json` in the usual mcpServers shape, with each server's command, args and env. For S3-driven master mode, `S3_BUCKET_NAME`, `S3_ACCESS_KEY`, `S3_SECRET_KEY` and `S3_REGION`; for the package registry, `MCP_FETCH_URL` and `MCP_TOKEN`. Installed packages land under `.fmcp-packages/Author/Package/Version/` with a metadata.json. Licensed GPL-3.0.

Setup effort

One command plus a key — pip install fluidmcp, then supply credentials