Librarian MCP Server

What is Librarian MCP Server?

Librarian MCP Server is a read-only Model Context Protocol (MCP) server that provides an API for listing, searching, and retrieving markdown files stored in a structured directory hierarchy. It is designed to act as a knowledge base for Large Language Models (LLMs), delivering document content on demand through the MCP framework.

How to use Librarian MCP Server?

Install Node.js v14+, clone the repository, run pnpm install then pnpm build. Configure the document root via the CLI argument --docs-root or the environment variable LIBRARIAN_DOCS_ROOT. Start the server with node dist/bin.js. Integrate with an MCP client to call tools such as getDocument, listDocuments, searchDocuments, and listTags.

Key features of Librarian MCP Server

• Structured document organization by section (e.g., daisyui/components/button.md)
• Tag-based filtering using frontmatter tags
• Hierarchical tag inheritance from parent index.md files
• Flexible search: case‑insensitive string or regex with configurable flags
• Efficient retrieval of one or multiple documents by path
• Tag discovery with usage counts and optional file paths
• Seamless MCP integration for LLM interaction

Use cases of Librarian MCP Server

• An LLM retrieving documentation for a specific component or section
• Searching document content with strings or regex patterns across a knowledge base
• Listing available topics and tags to understand the knowledge taxonomy
• Building context for LLM responses by fetching relevant document contents

FAQ from Librarian MCP Server

What dependencies are required?

Node.js v14 or higher and npm or pnpm.

Does Librarian support writing or modifying documents?

No. Librarian is a read‑only service and does not provide any writing operations.

How does tag inheritance work?

A document inherits all tags from index.md files in its parent directories. Tags are merged from the root (most general) to the document (most specific).

What search modes are available?

Two modes: "string" (case‑insensitive by default) and "regex" (with customizable case sensitivity and flags).

What error codes can the server return?

INVALID_ARGUMENT (invalid parameters), NOT_FOUND (missing document or directory), and INTERNAL (unexpected server error).