Skip to content
Sign inGet started
Sign inCreate account

MCP guides

On this page

Practical guides for teams that use MCP servers from Claude Code, Codex, Cursor and Claude Desktop. Each one starts with the answer, then covers setup, permissions and the limits worth knowing before you rely on it.

Choosing MCP servers

Controlling what agents can do

  • What is an MCP gateway?: how a gateway sits between AI apps and MCP servers, and when you don't need one. Start here; the guides below go deeper on each part.
  • MCP security risks: prompt injection, tool poisoning, token passthrough and the other attacks to plan for, with the control for each.
  • MCP authentication: how OAuth sign-in works for remote MCP servers, from discovery to token audiences.
  • MCP elicitation: how servers ask the person for input mid-call, in form and URL mode, and what clients must show.
  • MCP allowlist: limit which servers and tools an agent can use, in the client and at a gateway.
  • Claude Code permissions for MCP tools: permission modes, allow, ask and deny rules, and how they combine with gateway rules.
  • Claude Code --dangerously-skip-permissions: what bypass mode skips, what still applies, and how to keep MCP tools contained without prompts.

Comparing products and pricing

  • Composio pricing: Composio's plans, what counts as a tool call, add-on rates and what a small team pays.
  • MCP gateway alternatives: ten gateways and tool platforms compared by hosting, permissions and price.

Comparing MCP with other approaches

  • MCP vs CLI: when an agent should run gh or kubectl in a shell and when it should call an MCP server.
  • MCP vs Agent Skills: what SKILL.md skills and MCP servers each add, and how to combine them.
  • MCP vs A2A: agent-to-tool and agent-to-agent protocols, and how they fit together.
  • MCP vs API: how an MCP server can wrap a REST API for AI apps, and when to call the API directly.

Setting up your AI app