Most of what makes a new MCP server tedious is structural — process cleanup, error boundaries, logging to stderr rather than stdout — and a template gets that right without anyone thinking about it. The guidance tools cover the part a template cannot, which is the decision about what your server should expose in the first place. Saved workflows are worth using if you generate servers of a similar shape repeatedly.
A server for building servers. It generates a new MCP server project from a name, a description and a feature list, using a library of templates; alongside that it answers design questions with topic-specific guidance, and can save a multi-step creation process as a reusable workflow.
- `create_mcp_server` generates the project from `name`, `description`, `language`, `template_type` and a `features` list such as tools and resources
- `list_templates` shows what is available, filterable by language
- `get_ai_guidance` answers by `topic` and `server_type` — asking for security guidance on a database server returns advice shaped for that case
- `save_workflow` records an ordered set of steps, mixing input collection, guidance and generation, so the same process can be repeated
- A template library covering a plain Python server, one with resources for database and API integration, one using sampling, and a Gradio interface
- Templates are Jinja2 files with a metadata descriptor, so adding your own is a directory and two files
- A Gradio web interface for managing servers visually, as an alternative to driving it through a client
Python 3.10 or higher and the `uv` package manager. Clone the repository, create the environment and install it in editable mode, then copy the example environment file. AI guidance needs at least one model provider configured — `ANTHROPIC_API_KEY`, `OPENAI_API_KEY` or a local `OLLAMA_BASE_URL`. `DEFAULT_OUTPUT_DIR` decides where generated servers land, and `GRADIO_SERVER_PORT` controls the web interface. The client launches it through `uv` against the repository directory.