What Is MCP (Model Context Protocol)?
Anthropic's open standard for AI tool use — and why it needs governance.
Model Context Protocol (MCP) is an open protocol developed by Anthropic that standardizes how AI models connect to external data sources and tools. MCP provides a universal interface — rather than building custom integrations for each tool, developers implement one protocol. MCP servers expose capabilities; MCP clients (AI models) consume them.
How MCP Works
MCP defines a client-server architecture: MCP Servers expose tools, resources, and prompts through a standardized API. MCP Clients (AI models like Claude) discover and invoke these capabilities. The protocol handles authentication, capability discovery, and structured communication. A single MCP server can be used by any MCP-compatible model. A single model can connect to multiple MCP servers simultaneously.
The Security Challenge
MCP standardizes tool access, which also standardizes the attack surface. Every MCP server an agent connects to is a potential target for: unauthorized tool access, data exfiltration via tool calls, prompt injection through retrieved context, and privilege escalation through tool chaining. MCP provides the connection protocol. It does not provide governance over what actions are taken through those connections.
MCP + Exogram
Exogram ships a native MCP server that integrates directly with Claude Desktop and any MCP-compatible client. Every tool call that passes through the MCP server is evaluated by the deterministic policy engine before execution. This provides: schema validation for MCP tool calls, boundary control for MCP actions, loop protection for MCP-driven workflows, and full audit trail of every MCP interaction. The protocol provides the connection. Exogram provides the governance.
Beyond MCP: Universal Tool Governance
MCP is one protocol among many. OpenAI uses function calling. Anthropic uses tool use. Google uses function calling. Custom agents use REST APIs. Exogram governs all of them through the same execution boundary — one governance layer, every protocol, every framework, every model.
Frequently Asked Questions
Is MCP only for Claude?
No. MCP is an open protocol that any model can implement. While Anthropic developed it for Claude, MCP servers are model-agnostic — any MCP client can connect to any MCP server.
Does Exogram require MCP?
No. Exogram works with any tool-use mechanism — MCP, OpenAI function calling, Anthropic tool use, REST APIs, and custom integrations. MCP is one supported protocol, not a requirement.
How do I set up Exogram with MCP?
Exogram ships a native MCP server. Add it to your MCP client configuration (Claude Desktop, etc.) and every tool call is automatically governed by the deterministic policy engine. See the integration guide at /integrations/mcp.