Model Context Protocol (MCP)
An open standard that lets any AI application connect to external data and tools through one shared interface instead of custom code per system.
Category
Industry concept
See all terms
What is the Model Context Protocol?
The Model Context Protocol (MCP) is an open standard for connecting AI applications to external data, tools, and systems through one shared interface. An AI assistant that speaks MCP can reach any system exposing an MCP connection, with no integration built specially for that pairing.
Before MCP, every combination of an AI tool and a data source needed its own code. Connecting five assistants to five systems meant twenty-five separate builds, and each one broke on its own schedule. MCP replaces that with a single protocol both sides implement once.
How does MCP differ from a traditional API?
MCP is different from a traditional API integration: an API defines one service's own endpoints, so every caller writes bespoke code against it, while MCP defines a shared contract that any client and any system follow. Build an MCP connection once and every MCP-compatible tool can use it, with no work per tool. An API connects one program to one service, while MCP connects an ecosystem of AI applications to an ecosystem of data.
How does the Model Context Protocol work?
MCP runs on three roles. The host is the AI application the person uses, such as Claude, ChatGPT, or a coding tool. Inside it, a client holds one connection to one server. A server sits in front of a data source and exposes what that source offers: tools the model can call, resources it can read, and prompts it can run. The client and server negotiate a session, and from then on the model can list, search, read, and act through that connection while the server enforces what each caller may touch.
Qontext runs a hosted MCP server for the context repository. Any MCP client authenticates with an API key and gets a set of operations over company context: read with ls, find, and cat, search with grep and hybrid search, and write back through versioned Changes, all bounded by the user permissions and access controls set on each Client.