Claude Code 2.1.287: MCP server stopped connecting, what to set

If an MCP server stopped connecting after you updated to Claude Code 2.1.287, the changelog's own fix is to add "bareElicitationCapability": true to that server's MCP config entry. The release added URL prompts from MCP servers on the 2025-11-25 protocol, for example to sign in, and that setting is the escape hatch for servers that no longer connect after the change.

This is from the Claude Code changelog entry dated October 1, 2026, read 2026-10-01. Nothing in Sume's docs describes the hosted server sending URL prompts, so treat the setting as a fix to apply only if your Sume entry actually fails.

What did 2.1.287 change for MCP servers?

The changelog line reads: added URL prompts from MCP servers on the 2025-11-25 protocol, for example to sign in. It then gives the one-line remedy for a server that no longer connects. It does not say which servers are affected, so do not add the flag to entries that still work.

How do I tell that the Sume entry is the one that broke?

Confirm the entry first. Sume's quickstart registers the hosted endpoint with claude mcp add --transport http sume https://mcp.sume.com/mcp, then claude mcp login sume. A connection failure is not always the new setting, so run through the cheap checks in order.

Where does the flag go in the Sume entry?

The changelog says to add it to the server's MCP config entry. For the Sume remote server that is the entry in your .mcp.json or user config; the other keys stay as they were.

{
  "mcpServers": {
    "sume": {
      "type": "http",
      "url": "https://mcp.sume.com/mcp",
      "bareElicitationCapability": true
    }
  }
}

Does this change how I sign in to Sume?

No. The documented sign-in is OAuth: the client is sent to https://mcp.sume.com/oauth/authorize, then to the consent page on the MCP host, where Read is locked on and Write is an opt-in toggle. The flag concerns what the client advertises to a server, not Sume's consent page. If you use an API key instead, see OAuth and API keys.

What if it still fails after the flag?

Remove the flag again so you are not carrying an unneeded setting, and work through the sign-in and URL checks above. Retry every server that failed or needs authentication at once with /mcp reconnect all, which Claude Code added in 2.1.284.