What is MCP in AI?
MCP stands for Model Context Protocol, an open standard that lets AI applications connect to external tools and data through one common interface. An MCP server exposes tools, resources and prompts. An MCP client inside the AI app discovers and calls them. Anthropic published it, and it is now supported across many AI products.
The architecture is client-server. A host application such as Claude Desktop, Claude Code, Cursor or VS Code runs an MCP client. The client connects to one or more MCP servers over a transport, either standard input and output for a local process or streamable HTTP for a hosted service. On connect, the client asks each server what it offers. Tools are functions the model may call with typed arguments. Resources are read-only data the app can load into context. Prompts are reusable templates. The model never talks to the server directly: the host mediates every call, which is where permissions and approvals live.
JobsPipe's hosted server is a concrete example. It runs at mcp.jobspipe.dev over streamable HTTP and exposes four tools: search_jobs, which queries live postings with filters such as job_title_or, job_country_code_or, remote and skills_or; detect_company_tech_stack, which scans a domain; list_pricing_plans; and search_documentation, which returns matching sections of the API docs so the model can look up a filter name before using it. Adding it to Claude Code is one command, and any other client takes a short JSON entry naming the server URL. The client prompts for a JobsPipe sign-in the first time.
The practical effect is that a single integration serves every MCP-capable product. Before MCP, giving a model access to a data source meant writing a tool definition per framework and per app. With it, the server is written once, listed in a registry, and reused by every agent that speaks the protocol.
The full treatment is on MCP and AI agents.
Related questions
Where is MCP used?
In coding assistants such as Claude Code, Cursor, Windsurf and VS Code, in desktop chat apps such as Claude Desktop, in OpenAI's ChatGPT connectors and Agents SDK, and in agent frameworks that load tools from servers. Typical servers wrap source control, filesystems, databases, messaging tools and data APIs. JobsPipe's server gives any of those clients live job postings and tech-stack scans.
Is MCP only for Claude?
No. Anthropic published MCP as an open specification with open-source SDKs, and Claude products were the first clients. Since then OpenAI, Cursor, Windsurf, Microsoft's VS Code and others have added client support, so a server such as JobsPipe's works the same in each. Support varies by product and version, so check your client's documentation for which primitives it implements.
What is the difference between MCP and function calling?
Function calling is the model-level ability to emit a structured tool request. MCP is the plumbing that tells the application which tools exist, how to call them and how to reach the process that runs them. When a client connects to an MCP server, each tool becomes a function the model can call. MCP standardizes discovery and transport, not the model's reasoning.
Is MCP secure?
It is as secure as the server, the transport and the host's permission model. Remote servers should require authentication, the host should ask before running tools with side effects, and servers should declare what they do. JobsPipe's server requires a sign-in or bearer key, and all four of its tools are read-only lookups, so a connected agent can query data but never change anything.
More answers
Query live postings yourself, 1,000 jobs a month free.
Get a free API key