llmwiki-serve Released: Connect Obsidian Docs as Context for Coding Agents

llmwiki-serve, released by HwangJohn, is a local server that connects existing Markdown, Obsidian, and LLMWiki folders as read-only Knowledge Sources for coding agents (such as Codex and Claude Code). It focuses on retrieving document evidence via CLI, HTTP, and MCP without modifying original files, rather than acting as a wiki compiler.

Key Features and Scope

• Original Preservation: Reads wikilinks, YAML front matter, headings, tags, etc., in Markdown/Obsidian without modifying or uploading the originals.
• Various Interfaces: Provides the same data projection via CLI, HTTP, and MCP Streamable HTTP.
• Security Filtering: Documents marked as draft, unpublished, or confidential are excluded from default queries.
• Search Methods: Default lexical search requires no models or embeddings; Vector search must be explicitly enabled with the [vector] extra and FastEmbed provider.

Retrieval Action Guidance (System-One/Jev)

An optional feature added in version 0.2.13, disabled by default. After generating a context pack, it suggests next actions (stop, read more, re-search, etc.). Enable it using the LLMWIKI_QUERY_ACTION_JUDGE_API_KEY environment variable and the --query-action-judge system-one argument; only masked query information is passed to external providers.

Installation and Precautions

• Installation: uv tool install llmwiki-serve==0.2.13 (Python 3.11 or higher)
• Security: Binding to 127.0.0.1 is recommended for local use; authentication and TLS configuration are required for external exposure. Since I/O debugging logs may record queries/responses, disable them with --io-log off or configure a secure path.