- EOF2025 notes
- AI Engineer Summit 2025
- Anthropic MCP workshop
- kinda like language server for model context
- external systems
- prompts
- pre-defined prompt templates
- can be interpolated with data from application and user
- tools
- model-controlled - LLM chooses when to invoke tool
- can be retrieval, updating, anything
- resources
- data exposed to the application
- e.g. files, database records, API responses
- client can subscribe to updates from server
- MCP avoids the need to copy paste from external resources
- can build MCP once, be adopted everywhere
- MCP could be the “foundational protocol” for agents
- Recently added support for remote MCP servers
- Registry
- No centralized way to get servers
- Would be nice if there was a registry to enable discovery of remote MCP servers
- Wandb workshop
- Presenter used reflect.app for notes - looks interesting
- Letta workshop
- Cool workshop with a Jupyter Notebook for everything demonstrated
- At the end there was a demo of two LLM agents talking to each other