Summary
When a user runs /mcp in the Claude Code CLI to reconnect an MCP server (e.g., Slack), it only refreshes the local OAuth token. The remote connector registered at claude.ai/customize/connectors — used by cloud routines (CCR) — is a separate OAuth registration and is not updated.
This means users who reconnect an MCP locally still have broken cloud routines until they separately navigate to claude.ai/customize/connectors and reconnect there too.
Expected behavior
/mcp reconnect (or an equivalent CLI command) should also refresh the corresponding remote connector token so both local sessions and cloud routines stay in sync.
Current behavior
- Local:
/mcp refreshes token → local session works
- Remote: cloud routine connector token remains stale →
invalid_auth_token errors in scheduled routines
Steps to reproduce
- Have a Slack MCP connector set up both locally and as a remote connector for a cloud routine
- Let the Slack token expire
- Run
/mcp to reconnect Slack in the CLI
- Trigger a cloud routine that uses Slack → still fails with
invalid_auth_token
- Must separately go to
claude.ai/customize/connectors to fix the remote connector
Why this matters
Users operating scheduled cloud routines depend on remote connectors staying valid. When a token expires, the fix is non-obvious (two separate places to reconnect) and the failure is silent — the routine fires on schedule but produces no output.
Suggested fix
Link local MCP reconnections to their corresponding remote connector registrations, or at minimum surface a warning in the CLI when a local reconnect doesn't cover a paired remote connector.
Summary
When a user runs
/mcpin the Claude Code CLI to reconnect an MCP server (e.g., Slack), it only refreshes the local OAuth token. The remote connector registered atclaude.ai/customize/connectors— used by cloud routines (CCR) — is a separate OAuth registration and is not updated.This means users who reconnect an MCP locally still have broken cloud routines until they separately navigate to
claude.ai/customize/connectorsand reconnect there too.Expected behavior
/mcpreconnect (or an equivalent CLI command) should also refresh the corresponding remote connector token so both local sessions and cloud routines stay in sync.Current behavior
/mcprefreshes token → local session worksinvalid_auth_tokenerrors in scheduled routinesSteps to reproduce
/mcpto reconnect Slack in the CLIinvalid_auth_tokenclaude.ai/customize/connectorsto fix the remote connectorWhy this matters
Users operating scheduled cloud routines depend on remote connectors staying valid. When a token expires, the fix is non-obvious (two separate places to reconnect) and the failure is silent — the routine fires on schedule but produces no output.
Suggested fix
Link local MCP reconnections to their corresponding remote connector registrations, or at minimum surface a warning in the CLI when a local reconnect doesn't cover a paired remote connector.