MCP (Model Context Procotol)
- — What is MCP?
- — Giving tools and information to model so we can give it context about us and tools to execute
- — As long as an AI component provide an interface to MCP, any piece of software can talk to any others
- — Solves the M x N problem
- — Primitives
- — Tools
- — Resources
- — Prompts
- — Components
- — Server
- — An independent process in host's computer or remote server that provides tools and resourses and prompts
- — Give code and configuration to an SDK and then you can run a command to start a process
- — Client
- — In the host, for each connection to a server, a new client is created
- — Manages the protocol level things
- — Serializing and deserializing
- — Getting a list of tools availabing by querying
- — List of clients is given to host in configuration file as a JSON file
- — Can give a SSH link or the command to run locally
- — Host
- — Any someware that uses AI models and wants to connect to MCP services
- — Has access to an LLM
- — Gets the inputs from users, gets information from MCP servers and coordinates the messages with LLM
- — If LLM wants to make a tool execution, it will give information to host, host calls the tool through MCP
- — List of MCP given to
- — How to identify an MCP
- — From a remote URL
- — The command, with a path to file, to call and run the process locally
- — Here the STDIO is used to communicate
- — SDK