LLMling

What is LLMling?

LLMling is a Python framework for declarative LLM application development focused on resource management, prompt templates, and tool execution. It provides a YAML-based configuration system and serves as the backend for an MCP server and a pydantic-AI based Agent.

How to use LLMling?

Use the CLI to manage configurations and interact with components (llmling config init, llmling resource list, llmling tool call, llmling prompt show). Integrate with agents via Python code using RuntimeConfig and LLMlingAgent. Start an MCP server for editors like Zed or Claude Desktop using uvx mcp-server-llmling@latest start.

Key features of LLMling

• Declarative YAML-based configuration for LLM applications
• Built on the Model Context Protocol (MCP)
• Supports Resources, Prompts, and Tools as component types
• 100% typed Python with pydantic(-ai) foundation
• File watching with .gitignore-style patterns
• Resource processing pipeline through custom processors

Use cases of LLMling

• Define an LLM's environment entirely in YAML without code
• Load and pre-process content from files, URLs, CLI output, or Python callables
• Create structured prompt templates with arguments for consistent interaction
• Extend LLM capabilities with Python functions as tools
• Run an MCP server for integration with editors like Zed or Claude Desktop

FAQ from LLMling

What runtime does LLMling require?

Python 3.12 or higher is required; the package is 100% typed and pydantic(-ai) based.

What is the configuration format?

LLMling uses a YAML-based configuration system where you define resources, prompts, and tools declaratively.

How does LLMling integrate with editors?

LLMling can be used as an MCP server with Zed Editor (via settings.json) or Claude Desktop (via claude_desktop_config.json).

Can resources be dynamically updated?

Yes, resources like path and image types support file watching with configurable patterns and exclusion rules.

What are resource groups?

Resource groups allow you to group related resources for easier access, such as grouping together resources used for code review.