[BUG] Routines: claude.ai MCP connectors (Gmail, Slack, Notion) not injected in remote session

Resolved 💬 4 comments Opened May 21, 2026 by michigarciadiez-hub Closed Jun 22, 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?

Connectors configured in the Routine UI panel (Gmail, Slack, Notion)
are not loaded as MCP tools when the routine executes in remote session.
The model cannot find any of the configured connector tools and responds
as if no connectors are available.

What Should Happen?

Connectors enabled in the Routine panel should inject their MCP tools
into the session automatically on startup, without requiring a user
message to warm the session.

Error Messages/Logs

API Error: 400 messages.32.content.1.text: 
cache_control cannot be set for empty text blocks

Model response during execution:
mcp__gmail__search_threads — No disponible
mcp__slack__post_message — No disponible
mcp__notion__create_page — No disponible

Steps to Reproduce

  1. Connect Gmail, Slack and Notion at claude.ai/settings/connectors
  2. Create a Routine at claude.ai/code/routines with those three

connectors enabled in the connector panel

  1. Add a prompt that calls Gmail search_threads as first action
  2. Execute the routine (manual trigger)
  3. Observe: model reports all MCP tools as unavailable

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Last Working Version: Deja vacío (no sabes cuándo funcionó)

Claude Code Version

Web (claude.ai/code/routines) — no CLI version available

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

Related issues: #43397, #47383, #22726
Same connectors work correctly in normal interactive claude.ai chat sessions.
Only fails in Routines remote execution context.

View original on GitHub ↗

This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗