Glossary
MCP
An open protocol from Anthropic that standardizes how language models discover and call external tools, data sources, and prompts via a small JSON-RPC interface.
A protocol layer between AI applications and the data, tools, and prompts they need. An MCP server exposes a set of capabilities (resources, tools, prompts) over JSON-RPC; an MCP client (the host application embeddingretrieval-memoryA fixed-size vector representation of a piece of text learned so semantically similar texts land near each other in the vector space, the basis for vector search and most RAG. Open full entry the model) discovers and invokes them. The point is to make tool integrations portable across models and across applications.
Anthropic released the initial spec in late 2024; through 2025 the Linux Foundation took on stewardship, OpenAI and Google added client- side support, and an ecosystem of community servers emerged for filesystems, GitHub, databases, Slack, browsers, etc.
The competing approach is provider-specific tool schemas. MCP’s pitch is that one tool implementation works regardless of which model is asking. The tradeoff is an additional protocol layer to operate and secure: confused-deputy attacks and prompt injectionsafety-guardrailsAn attack where adversarial content in a document, tool result, or web page is interpreted as instructions by the model, overriding the user or system prompt. Open full entry through MCP tools have been the main reported vulnerabilities in 2025.