Add automatic MCP server health check on session start

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Summary

There's currently no way to automatically check and notify users about disconnected MCP servers when a session starts. MCP servers using OAuth authentication can silently disconnect when tokens expire, and users only discover this when a tool call fails mid-workflow.

Problem

• MCP servers using OAuth flows can disconnect between sessions due to token expiry
• Users have no way to know which servers are disconnected until they try to use them
• Hooks (SessionStart) can run shell commands but cannot check MCP connection status or reliably force the AI to report on it
• No CLI command exists to query MCP server connection state (e.g., claude mcp status)

Proposed Solution

One or more of the following:

1. Auto health check setting — A setting like "mcpHealthCheck": ["my-mcp1", "my-mcp2"] that checks specified MCP servers on session start and notifies the user of any that are disconnected.
2. claude mcp status CLI command — A command that returns connection status for all/specified MCP servers, usable from hooks.
3. Built-in session start notification — Automatically display disconnected MCP servers to the user when a session begins, similar to how tool availability is shown.

Use Case

Teams sharing Claude Code configurations across multiple developers with OAuth-based MCP servers. Currently there's no scalable, reliable way to surface connection issues at session start.

Alternative Solutions

1. SessionStart hook with directive message — Used a hook to echo a message instructing the AI to check MCP tool availability and report disconnected servers. Unreliable because the AI doesn't always act on hook output.

1. Persistent memory (MEMORY.md) — Saved instructions in the AI's memory to check MCP servers on startup. User-specific and doesn't scale when sharing configurations across teams.

1. Manual /mcp check — Running /mcp manually at the start of every session. Works but defeats the purpose of automation.

None of these approaches are reliable or scalable. A built-in solution is needed.

Priority

High - Significant impact on productivity

Feature Category

CLI commands and flags

Use Case Example

_No response_

Additional Context

A team uses Claude Code with OAuth-based MCP servers like Jira and an internal service. OAuth tokens expire periodically. When a developer starts a new session:

1. They begin working and ask Claude to fetch a Jira ticket
2. The tool call fails because the OAuth token expired overnight
3. They run /mcp, re-authenticate, and retry
4. This cycle repeats across sessions and team members

Expected behavior: On session start, the user is automatically notified: "MCP servers disconnected: jira". Run /mcp to re-authenticate." — before they hit a failure mid-workflow.