MCP Servers
The open registry for Model Context Protocol servers. Find the right tools, resources, and prompts for your AI agents — filtered by category, transport, or use case.
Servers
206
Tools
545
Categories
11
Contributors
182
The official Weights & Biases MCP server for W&B Models and Weave, letting AI assistants query and analyze experiment tracking data in natural language. Agents can look up runs and their metrics, compare sweeps, inspect registered model artifacts, and dig into Weave traces to debug LLM application behavior.
A self-contained MCP server for managing local Ollama models with no external dependencies. Agents can list installed models, pull or remove them, start and stop the Ollama server, run local completions, and use a model-advisor tool that recommends which locally available model best fits a given task.
Skills vs MCP servers
what’s the difference?Skillsthe “what to do”
A skill packages know-how — instructions, an example prompt, and recommended models — so an agent performs a task consistently. Skills add knowledge, not new connections.
MCP serversthe “how to connect”
An MCP server gives an agent new capabilities by connecting it to real systems (databases, APIs, files) over a transport. MCP adds connections and actions, not task instructions.
Rule of thumb: reach for a skill when you need the model to do a task well, and an MCP server when you need it to reach a tool or system. They compose — a skill can rely on tools an MCP server provides.
Built an MCP server?
Submit it to the registry — it’s open source and community-maintained.