Agent Shelf vs LangChain Hub

Overview

Agent Shelf and LangChain Hub both let you share reusable AI prompts and agent definitions, but they serve different audiences. LangChain Hub is a prompt repository tied to the LangChain Python/JS framework. Agent Shelf is a tool-agnostic registry where agents work across any AI coding tool.

Key differences

Target audience

Agent Shelf is built for anyone using AI coding tools — developers, marketers, writers, DevOps engineers. You don't need to write code to use an agent. Just download the Markdown file and use it with your preferred tool.

LangChain Hub is built for developers who write LangChain applications in Python or JavaScript. Using a prompt from the Hub requires writing LangChain code to load and execute it.

Agent definition format

Agent Shelf uses Markdown files with YAML frontmatter. The format is human-readable, version-controllable, and not tied to any framework. An agent's instructions, metadata, and tool declarations are all in one portable file.

LangChain Hub stores prompt templates in LangChain's format. Prompts are structured for LangChain's chain/agent abstractions and may include template variables, output parsers, and chain configurations specific to LangChain.

Framework independence

Agent Shelf agents work with Claude Code, Cursor, Windsurf, GitHub Copilot, and any tool supporting the Agent Skills spec or MCP. No framework dependency.

LangChain Hub prompts require LangChain to run. You need langchain installed and write code like hub.pull("owner/prompt") to use them. Switching to a different framework means rewriting your prompt integration.

Built-in tool access

Agent Shelf agents run inside your development environment with direct access to your files, terminal, and project. They can declare MCP servers and skills that get auto-installed.

LangChain Hub prompts are text templates. Tool access depends on what you wire up in your LangChain code — the Hub itself doesn't handle tool configuration or installation.

Discovery

Agent Shelf offers web browsing, search, categories, tags, and an MCP server that lets your AI tool search and download agents programmatically — without leaving your IDE.

LangChain Hub has web-based browsing and search through LangSmith. Programmatic access is through the LangChain SDK.

Versioning and history

Agent Shelf uses semantic versioning (semver). Each publish creates an immutable snapshot. Users can download any specific version or always get the latest.

LangChain Hub supports commit-based versioning. Prompts have a commit history, and you can pull specific commits by hash.

When to choose each

Choose Agent Shelf when:

• You use AI coding tools (Claude Code, Cursor, Windsurf, etc.)
• You want agents that work across multiple tools
• You want to share agents with non-developers
• You prefer human-readable Markdown over framework-specific formats
• You need agents with bundled tool integrations

Choose LangChain Hub when:

• You're building a LangChain application in Python or JavaScript
• You need prompt templates with variables and output parsers
• You want tight integration with LangSmith for tracing and evaluation
• Your team is already standardized on the LangChain ecosystem

Feature comparison

| Feature | Agent Shelf | LangChain Hub | |---------|------------|---------------| | Format | Markdown + YAML | LangChain prompt templates | | Framework required | None | LangChain (Python/JS) | | Works with | 40+ AI coding tools | LangChain apps only | | IDE integration | Native (runs in your tool) | Requires custom code | | Versioning | Semantic versioning | Commit-based | | Tool bundling | MCP servers, skills, deps | Not supported | | Discovery | Web + MCP server | Web + SDK | | Price | Free | Free (LangSmith has paid tiers) | | Private sharing | Per-user permissions | Organization-based |