# Model Context Protocol (MCP) ## Summary Model Context Protocol (MCP) is a system interface that allows language models to retrieve external context and take structured actions through API-connected services during inference. ## Definition MCP is an open protocol for extending the operational scope of large language models (LLMs) beyond static prompts. It standardizes how tools and services can: 1. Offer real-time, structured data to models, 2. Interpret model outputs as executable actions, 3. Maintain a traceable, secure bridge between decision contexts and downstream effects. While the model itself remains an autoregressive token predictor, MCP enables *workflow agency*—giving the appearance of reasoning by collapsing the gap between decision context and operational execution. This shifts LLMs from isolated completion engines to dynamic interface layers in human-machine systems. ## Properties - **Structured context input**: Models can now ingest dynamic data from platforms like Salesforce, Airtable, or Slack in standardized formats mid-inference. - **Action execution layer**: Through partners like Zapier, model outputs can become system-level actions (e.g. updating records, sending messages, triggering automations). - **Integration-first design**: MCP is not a model update; it is a protocol layer that makes LLMs interoperable with real-time systems. - **Orchestration-enabler**: MCP doesn't make the model more intelligent—it removes bottlenecks around execution, increasing the perceived intelligence. ### Wayfinding %% DATAVIEW_PUBLISHER: start ```dataview TABLE WITHOUT ID questions AS "Responds to", origins AS "Informed by", ideas AS "Developed alongside", concepts AS "Builds on" WHERE file.name = this.file.name