What is an MCP Client?
An MCP client is the component within an AI agent or application that connects to MCP servers, discovers available tools and resources, and invokes them on behalf of the agent following the Model Context Protocol specification.
WHY IT MATTERS
In the MCP architecture, the client is the agent's gateway to external capabilities. It maintains connections to one or more MCP servers, each providing different tools, resources, or prompts. The client handles protocol negotiation, capability discovery, and request/response management.
Most agent frameworks and tools now include MCP client support — Claude Desktop, Cursor, Windsurf, VS Code Copilot, LangChain, and OpenAI Agents SDK all speak MCP natively. This means any MCP-compatible tool automatically works with any of these clients.
The client is typically unaware of what sits between it and the server. This is what makes proxy-based policy enforcement possible — the client connects to what it believes is the server, but is actually a policy-enforcing proxy.
HOW POLICYLAYER USES THIS
PolicyLayer is fully transparent to MCP clients. The client connects to PolicyLayer exactly as it would connect to any MCP server — same protocol, same capability discovery, same tool invocation. No SDK, no plugin, no code changes. The client cannot distinguish PolicyLayer from the upstream server, which means policy enforcement works with every MCP client out of the box.