nabchan-mcp-server

What is nabchan-mcp-server?

nabchan-mcp-server is an experimental MCP server that returns information about the Nablarch framework based on its official documentation. It uses DuckDB and Lindera for full-text search of the documentation content.

How to use nabchan-mcp-server?

You can run it using Docker (with stdio transport) or via ToolHive. The README provides VSCode configuration examples for both stdio and SSE transports. For development, set up Python 3.11, uv, Git, and Docker.

Key features of nabchan-mcp-server

• read_document – Converts a Nablarch documentation URL to markdown.
• search_document – Searches the documentation and returns title, URL, and summary.
• Index built from Nablarch HTML docs using morphological analysis.
• Runs locally with Python and DuckDB.
• Supports stdio and SSE transport types.

Use cases of nabchan-mcp-server

• Ask questions about Nablarch in GitHub Copilot Chat (Agent mode).
• Search for specific Nablarch topics or components.
• Retrieve markdown versions of official Nablarch documentation pages.
• Integrate Nablarch documentation queries into AI‐assisted development tools.

FAQ from nabchan-mcp-server

Is the server production‑ready?

No, this is an experimental project with room for improvement.

What are the system requirements for development?

Python 3.11, uv, Git, and Docker are required. The index must be built using uv run -m tools.build_index.

Where does the documentation data come from?

Data is extracted from the official Nablarch documentation HTML files. Text is extracted with BeautifulSoup, summarized (via LLM), and indexed in DuckDB with Lindera for full‑text search.

What transport methods are supported?

Both stdio (default for Docker) and SSE (with a pre‑started HTTP server on localhost:8000) are supported.

How can I run the server without Docker?

Use ToolHive (thv run) or run directly with uv: uv run -m nabchan_mcp_server.main (with optional --transport sse and --host localhost).