The MCP Client Integration Prompt

Help me integrate MCP servers into my agent application.

Application: [DESCRIBE]
Servers to connect: [LIST — MINE, THIRD-PARTY, OR BOTH]
Trust level of each: [DESCRIBE]

Produce:
1. Connection lifecycle: when servers connect, reconnect, and
   disconnect
2. Tool discovery: how the agent learns what is available
3. Namespacing: preventing name collisions across servers
4. What to do when a server is unavailable mid-run
5. How server tools are presented to the model alongside native
   tools
6. Configuration: how users add, remove, and configure servers
7. Credential handling per server

SECURITY — address these specifically:
8. A third-party MCP server is untrusted code with a channel into
   your model's context. What are the risks?
9. Tool descriptions from a third-party server are text the model
   reads and acts on. How do you handle a server whose tool
   description contains instructions?
10. What should require user approval before a third-party server's
    tool runs?
11. How do you limit what a third-party server can reach?

Mark protocol specifics as [VERIFY AGAINST CURRENT SPECIFICATION].