zk‑mcp: Bridging Plain‑Text Notes and Large Language Models

zk‑mcp is an MCP (Model Context Protocol) server that turns a plain‑text note repository managed by zk into a fully queryable knowledge base for AI assistants. By exposing the core operations of zk—search, link analysis, tagging, and note creation—as MCP tools, it allows language models to read from, write to, and navigate a user’s personal notes with the same ease as calling any other external API. This integration removes the friction of manual file handling and enables assistants to surface relevant context, generate new content, or update existing notes directly from conversational prompts.

The server offers a rich set of tools that mirror the most common tasks developers and writers perform when working with zk. lets an assistant filter notes by content strings, tags, or exclusion criteria, returning paths that match complex logical expressions. exposes the bidirectional graph of note relationships, revealing which notes a target links to or is linked from, as well as related topics. provides a catalog of all tags for quick filtering, while retrieves the full text of a specific note. Finally, allows the assistant to add new entries programmatically, enabling dynamic knowledge expansion without leaving the chat interface.

In practice, zk‑mcp empowers a variety of real‑world scenarios. A developer can query their technical notes to surface code snippets or design decisions during a code review, while a researcher can pull related literature from their personal bibliography stored in zk. Writers can generate drafts that reference existing plot points or character details, and the assistant can automatically create new notes for emerging ideas. Because all operations are exposed through MCP, these workflows integrate seamlessly into any AI toolchain—Claude Desktop, Continue, or custom LLM pipelines—without the need for bespoke adapters.

What sets zk‑mcp apart is its lightweight, type‑safe implementation. Built on FastMCP and Pydantic, every tool returns well‑structured JSON, ensuring reliable parsing by the client. The server delegates actual note manipulation to the native zk CLI, preserving zk’s proven performance and offline capability. This design keeps the MCP layer thin, reduces latency, and guarantees that updates to zk’s storage format automatically propagate to the assistant without additional maintenance.

For developers who already use zk as a personal knowledge base, zk‑mcp turns that collection into an interactive, AI‑ready resource. It bridges the gap between static notes and dynamic conversational agents, enabling richer, contextually aware interactions that scale with your growing body of knowledge.