Model Context Protocol (MCP)

• Standardizes how AI systems (like LLMs) connect with external tools and data sources
• Launched by Anthropic in December 2024
• It's connector between the LLM models and the context in which it needs to operate (the outside world)
• Key features
• Portability: You don't need to rewrite integrations for each LLM vendor.
• Security: It standardizes permissions and sandboxing, so agents don't get unrestricted access.
• Ecosystem: Third-party developers can build MCP "servers" (tools, connectors) that any AI app can use.

Architecture

• MCP Client: The AI model or chat interface (Claude Desktop, VS Code plugin, etc.).
• MCP Server: The external tool/service you want the AI to access (e.g., GitHub repo, Postgres DB, Jira).
• MCP Protocol: Defines how the client and server exchange messages (structured JSON).

Servers

• Databases
• APIs
• Filesystems: let's AI read/write project files
• Dev tools (like VS Code, GitHub, CI/CD)
• Other agents

VSCode

-