/Glossary · AI
What is Model Context Protocol (MCP)?
Model Context Protocol (MCP) is an open standard, introduced by Anthropic, that defines how AI applications connect to external tools, data, and services. It lets you write an integration once and use it from any compatible client.
How MCP works: clients, servers, and transports
MCP uses a client-server architecture over JSON-RPC. The host application, such as Claude Code or your own agent, creates one client per MCP server. Each server exposes standard capabilities: tools the model can call, resources with context data, and reusable prompts.
The specification defines two transports. With stdio, the client launches the server as a local process and they talk over standard input and output. With Streamable HTTP, the server runs as a remote service and receives messages over HTTP, so it can be shared across users and deployed in containers or on Lambda.
Security for MCP servers
For remote servers, the specification defines an authorization framework based on OAuth 2.1; local stdio servers take their credentials from the environment. An MCP server has exactly the access you give it, so apply least privilege, tightly scoped IAM roles, and read-only operations whenever they are enough.
Also account for prompt injection: content returned by a tool can carry malicious instructions. Log every invocation, validate inputs, require human confirmation for destructive actions, and only use servers from known sources with pinned versions.
MCP use cases on AWS
AWS publishes open-source MCP servers for searching its documentation, working with services like CloudWatch or Cost Explorer, and generating infrastructure as code. Connected to Claude Code, they let an agent check real metrics, costs, or resources before proposing a change.
It also makes sense to build your own servers that expose internal APIs, runbooks, or a read-only database replica. Deployed on ECS or Lambda behind authentication, they give your agents on Amazon Bedrock controlled access to company systems.
/FAQ
Questions about Model Context Protocol (MCP)
Another question? Write to hello@nubent.com and an engineer will answer.
Does MCP only work with Claude?
No. MCP is an open standard with official SDKs in several languages, and clients and platforms from different vendors have adopted it. The same MCP server can be used from Claude Code, from your own agent on Amazon Bedrock, or from any other application that implements the protocol.
What is the difference between MCP and a REST API?
A REST API exposes operations to any program. MCP adds a layer designed for models: the server describes its tools with schemas and descriptions the agent reads to decide what to call, and the protocol standardizes discovery, execution, and authorization. Many MCP servers wrap existing REST APIs.
How do I deploy an MCP server on AWS?
For shared use, deploy the server with the Streamable HTTP transport on ECS, Lambda, or Amazon Bedrock AgentCore, behind OAuth authentication and with a least-privilege IAM role. For individual use, a local stdio server launched by the client itself is enough.
How we help with Model Context Protocol (MCP)
MCP servers
Custom MCP servers that connect Claude and other agents to your internal systems and AWS, with authentication, least privilege, audit logging and deployment in your account.
Learn moreAI Agents with Bedrock
Agents that carry out real tasks with Amazon Bedrock and AgentCore: tool use, MCP, Knowledge Bases, multi-agent orchestration, identity and observability.
Learn moreClaude Code for teams
Claude Code adoption for engineering teams: setup, CLAUDE.md, permissions, MCP, CI/CD, code review, credential security and measuring the impact on delivery.
Learn moreAI training
AI workshops for engineering teams: Claude Code, agents and RAG on Amazon Bedrock, MCP servers and AI security, with exercises built on your own codebase.
Learn moreNext step
What should we build next on AWS?
30 minutes with an engineer. Concrete next steps, whether or not you work with us.