The Model Context Protocol (MCP) is an open standard that defines how AI applications connect to external tools, data sources and systems through a common client-server interface, so that one integration works across any compatible AI model or assistant.
Anthropic introduced the Model Context Protocol in November 2024. On 9 December 2025, Anthropic donated it to the Agentic AI Foundation, a fund of the Linux Foundation co-founded with Block and OpenAI. By then MCP counted more than 97 million monthly SDK downloads and 10,000 active servers, with client support in ChatGPT, Claude, Gemini, Microsoft Copilot and Visual Studio Code.
MCP defines three roles and three primitives:
- Host: the AI application in which the user works
- Client: the connector inside the host that keeps a session open with one server
- Server: the program that exposes a system, such as an ERP, a bank connection or a document store
- Tools, resources and prompts: the actions a server lets the model execute, the data it lets the model read, and the templates it provides
For finance, MCP moves the integration question from whether an assistant connects to the ERP to what that connection is allowed to do. An MCP server that can read ledger data and also post journal entries or release payments gives an AI model the same power as a user with those rights. Scopes, approval steps and logging become the real design decisions, and servers that pass untrusted documents to a model add exposure to prompt injection.
Phacet applies the same principle to its agents: each agent works within defined permissions, any change to financial data goes through human-in-the-loop approval, and every action is logged in the audit trail. Hosting and data handling are detailed on the security page.
