The gap it fills is between hand-written Python and a help-text scraper: a typed, version-controlled mapping from your CLIs to MCP tools, with a validate command so a broken config fails in CI rather than in someone's editor.
A runtime that turns a configuration file into a running MCP server. You declare each tool — the command, the subcommand, the typed arguments and how output is captured — and one command serves them. Where other approaches scrape a binary's help text, here the mapping from CLI to tool is structured and typed, which makes it reviewable in a pull request and checkable in CI.
- `teukhos serve` — starts the MCP server from the config, with `--transport` overriding stdio or http and `--port` overriding the default 8765
- `teukhos install` — registers the server with your MCP clients: `--client` targets one by slug, `--all` covers every client detected, `--project` writes a project-level config, and `--dest` writes to any JSON file with `--config-key` picking `mcpServers` or `servers`
- `teukhos discover` — generates a config from a binary's own `--help`, recursing into subcommands to `--max-depth`, with `--filter` to limit it to one prefix and `--dry-run` to print rather than write
- `teukhos validate` — exits 0 or 1 against a config, which is what makes it usable as a CI check, and `teukhos wait-ready` polls until the server answers
- `teukhos clients` to see which clients are supported and which are installed, and `teukhos uninstall` to remove a registration
`pip install teukhos`, or run it through `uvx`. Configuration lives in `teukhos.yaml` by default, and the config path is passed explicitly on the commands that take one. For HTTP mode the `--url` and `--key` options on install point clients at a remote server, with the key defaulting to an environment reference rather than a literal. The project describes itself as beta: the core is stable and tested, and the API may change before v1.0.
One command — pip install teukhos