What is Librarian and what problem does it solve?

Librarian is an intelligent MCP skills server that manages and serves AI-ready knowledge skills. It provides fine-grained, secure, and verifiable access to knowledge for AI models, optimizing context windows and preventing prompt injection.

How does Librarian ensure the security and integrity of knowledge skills?

It employs robust security measures including Ed25519 digital signatures, SHA-256 manifests for tamper-proof verification, HMAC authentication, Unix domain sockets, and a content guard to prevent prompt injection. It also has zero external dependencies for minimal supply chain risk.

Which AI development environments or clients can use Librarian?

Librarian is designed to work with any MCP-compatible client that speaks MCP stdio, including popular tools like Claude Code, cclocal (Ollama), Crush, and Aider, integrating seamlessly into existing AI workflows.

What advanced search and retrieval capabilities does Librarian offer?

Librarian features BM25 search with sub-section chunking and suffix stemming for highly relevant, fine-grained content retrieval. It also utilizes progressive disclosure, delivering truncated content snippets with the option to load full sections to optimize AI context windows.

Librarian: Secure AI Knowledge Skills Server for MCP Clients

Librarian is an intelligent, zero-dependency MCP skills server designed to efficiently deliver context-specific knowledge to AI models. It features a robust BM25 search engine with suffix stemming and sub-section chunking, ensuring models receive precisely the information they need, typically small, relevant chunks. With Ed25519 digital signatures and SHA-256 manifests, it guarantees the integrity and authenticity of all skills. The server supports progressive disclosure, offering truncated content with an easy path to access full sections, keeping AI context windows lean. Built for any MCP-compatible client like Claude Code, cclocal, Crush, or Aider, Librarian also provides strong security measures, including Unix domain sockets, HMAC authentication, and a content guard to prevent prompt injection while allowing technical payloads.

Key Features

01BM25 search with suffix stemming and sub-section chunking for fine-grained retrieval.

02Zero external dependencies, ensuring a minimal supply chain risk and robust security footprint.

03Robust security architecture including Unix domain sockets, HMAC challenge-response authentication, and a content guard for prompt injection prevention.

040 GitHub stars

05Ed25519 digital signatures and SHA-256 manifest for tamper-proof skill integrity verification.

06Progressive disclosure of skill content, returning truncated chunks with an option to load full sections.

Use Cases

01Serving context-specific knowledge chunks to AI models for efficient task execution and problem-solving.

02Distributing and managing curated knowledge skill packs across development teams or organizations.

03Creating and curating new skills with standardized formatting, templating, and AI-assisted tooling.