model context protocol: anthropic’s open source standard specifying how AI models interact with external tools
https://www.figma.com/resource-library/what-is-mcp/
- 2-way communication protocol
- data providers and app devs who want LLMs to be able to access their data or do action → build MCP server
- devs building AI tools → MCP client enabling AI models to connect to MCP servers to fetch context or do tasks in other apps
- client-server architecture
- example:
- Cursor could have MCP client to connect to Figma and Github’s MCP server
benefits
- consistent universal interface instead of custom integrations with diff external apps
- lets AI assistants use one set of commands for all APIs
- inconsistent API definitions may be missing data or difficult for LLMs to interpret
difficulties
- not designed for scalability - a connection requires a port, and could live indefinitely
- not inherently fault-tolerant