Remote Trigger MCP connectors not injected into CCR session

Resolved 💬 3 comments Opened Apr 4, 2026 by cnapz Closed Apr 10, 2026

Summary

MCP connectors (Notion, Supabase) configured on Remote Triggers are not being injected into the CCR (Claude Code Remote) session runtime. The connectors show as connected in the trigger config and at claude.ai/settings/connectors, but the remote agent session cannot discover them via ToolSearch.

Environment

  • Plan: Claude Code Max
  • Environment: Cynapse (env_01WXuTg37c38KQ4ozv9c64VX, kind: anthropic_cloud)
  • Affected triggers: 3 scheduled triggers (workflow-steward, self-improving-agent, proactive-agent)
  • MCP connectors configured: Notion (828d32a7-4c9c-48d0-b66f-b211015b5ed6), Supabase (1d8f8691-b1d6-49d3-81be-55bd2d3efe35)

Reproduction

  1. Create a Remote Trigger with mcp_connections including Notion and Supabase connectors
  2. Trigger fires on schedule (or via manual run API call)
  3. Inside the CCR session, the agent runs ToolSearch for Notion MCP tools — no tools discovered
  4. config/.env is not present (path: /home/user/cynapseOS/config/.env does not exist in CCR)
  5. No NOTION_API_KEY in environment variables
  6. Agent falls back to degraded mode — cannot read/write Notion databases

Expected Behavior

MCP connectors listed in trigger.mcp_connections should be available to the agent in the CCR session, just as they are in a claude.ai interactive session.

Evidence

The workflow-steward agent logged this in its degraded run (committed to git):

STATUS: FAILED — NOTION UNREACHABLE

- Notion MCP: NOT CONNECTED (no tools discovered via ToolSearch)
- config/.env: NOT PRESENT in this environment
- NOTION_API_KEY: NOT SET in environment variables
- Notion API (direct HTTP): FAILED — no credential available

Root cause: This is the claude-code-on-the-go web instance. Notion MCP requires
either an injected OAuth connector or a configured MCP server. Neither is available
in this session.

Full log: https://github.com/cnapz/cynapseOS/blob/main/logs/agent-runs/workflow-steward-2026-04-04T02-23-38Z.log

Timeline

  • 2026-03-29: Triggers created with MCP connectors — worked initially
  • 2026-03-31: Last successful Notion write from a Remote Trigger
  • 2026-04-01 through 2026-04-04: All runs fail to reach Notion MCP
  • 2026-04-04: Re-authorized Notion at claude.ai/settings/connectors, refreshed mcp_connections via trigger update API — still not injected

Workaround

We added Supabase MCP fallback instructions to the trigger prompts so agents write to Supabase ct_agent_runs instead of Notion when MCP is unavailable. But the Notion connector should work as configured.

Impact

3 scheduled agents (running 5x/day total) cannot execute their primary functions. The Agent OS health monitoring, improvement cycle, and morning digest are all degraded.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗