[BUG] Custom/local MCP servers not available until after first response in Desktop App

Resolved 💬 2 comments Opened Apr 9, 2026 by haron1100 Closed Jun 25, 2026

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

Local/custom MCP servers are not available on the first message in a new Desktop App conversation. Only Anthropic-hosted connectors (ClickUp, HubSpot, Figma, etc.) load in time. Local MCP servers appear in the context (via <system-reminder> tags) only after the first response round-trip. Claude itself confirmed the deferred tool listings for local servers updated between the first and second message in the same session.

This does NOT occur in the CLI — only the Desktop App.

What Should Happen?

All configured MCP servers — both Anthropic-hosted connectors and local/custom servers — should be available from the very first message in a new conversation, as they are in the CLI.

Error Messages/Logs

No error messages. The MCP servers do eventually load — they just aren't available for the first message. Claude's first response lists only Anthropic-hosted connectors and states it has no access to the local MCP tools. By the second message, the same tools are fully available.

Steps to Reproduce

Steps to Reproduce

  1. Configure one or more local/custom MCP servers (e.g., Xero, or any local MCP server)
  2. Also have at least one Anthropic-hosted connector configured (e.g., ClickUp, HubSpot)
  3. Open a new conversation in the Claude Code Desktop App
  4. As the first message, ask: "What MCP servers do you have access to?"
  5. Claude will list only Anthropic-hosted connectors — local/custom MCP servers are missing
  6. Send a second message asking the same question
  7. Claude now correctly lists all MCP servers including local/custom ones

To confirm it's Desktop App specific:

  1. Open the Claude Code CLI in the terminal with the same configuration
  2. Ask the same question as the first message — all MCP servers (including local) are available immediately

Tested with multiple different local MCP servers — the issue is not specific to any one server.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.97

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Additional Information

macOS (Darwin 23.5.0)
Reproduced consistently across multiple new conversations
Tested with multiple different local/custom MCP servers — all exhibit the same behavior
The local MCP servers are fully connected and functional (not a connection failure) — they just resolve too late for the first message
The Anthropic-hosted connectors that DO load on the first message: ClickUp, Granola, Claude Preview, Claude in Chrome, Cloudflare, HubSpot, Figma, MCP Registry, Scheduled Tasks
Likely a race condition in the Desktop App's initialization pipeline where local MCP server resolution is async and doesn't complete before the first message is processed

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗