Fix MCP Inspector workflow - "Start MCP gateway" failure (80% failure rate)

[github-actions]

Problem

The MCP Inspector workflow is failing consistently with an 80% failure rate (8/10 recent runs failed). The workflow has been non-operational since 2026-01-05 (18 days ago).

Error Details

Failed step: Step 24 - "Start MCP gateway"
Last successful run: 2026-01-05
Recent failures:

• Run #37 - 2026-01-19T18:54:44Z
• Run #36 - 2026-01-16T19:24:44Z
• Run #35 - 2026-01-16T19:20:45Z
• Run gh add: better tracking of added/modified files #34 - 2026-01-12T18:55:15Z

Example run: §21148514645

Impact

• MCP tooling inspection capabilities are offline
• Cannot validate MCP server configurations
• Affects workflow development and debugging

Root Cause Analysis

The failure occurs at the "Start MCP gateway" step, which suggests:

1. Tavily MCP server configuration issue - The workflow uses Tavily MCP server (similar to Daily News which recently recovered)
2. MCP Gateway connectivity problem - Server may be unreachable or misconfigured
3. Missing or invalid secrets - May need TAVILY_API_KEY or other credentials

Related Context

Daily News workflow had similar TAVILY_API_KEY issue and recovered after the secret was added (2026-01-22). Research workflow is also failing with likely similar root cause.

Recommended Investigation Steps

1. Check Tavily configuration:

   • Verify TAVILY_API_KEY secret is accessible
   • Review .github/workflows/shared/mcp/tavily.md configuration
   • Compare with Daily News configuration (now working)

2. Review MCP Gateway logs:

   • Download artifacts from failed run: artifacts
   • Check /tmp/gh-aw/mcp-logs/ for error messages
   • Look for connection timeouts or authentication errors

3. Test MCP Gateway startup:

   • Run workflow manually with increased logging
   • Check if issue is specific to scheduled runs or all triggers
   • Verify Docker container images are accessible

4. Compare with working workflows:

   • Smoke Claude and Smoke Codex are healthy (90% success rate)
   • They don't use Tavily - check what MCP servers they use
   • Identify differences in configuration

Success Criteria

• MCP Inspector workflow runs successfully
• "Start MCP gateway" step completes without errors
• Success rate returns to >80% over next 5 runs

Priority: P1 (High)

This workflow provides critical tooling inspection capabilities. Fix within 24 hours to restore MCP debugging functionality.

References:

• Latest failed run §21148514645
• Workflow file

AI generated by Workflow Health Manager - Meta-Orchestrator

• expires on Jan 24, 2026, 2:59 AM UTC

[github-actions]

Status Update - 2026-01-24T02:51:00Z

MCP Inspector Status

• Still failing: 80% failure rate persists
• Latest failure: §21304877267 (2026-01-23)
• Days offline: 19 days (since 2026-01-05)
• Error: "Start MCP gateway" step continues to fail

Investigation Progress

Recent run shows workflow completed artifact uploads but still failed. This suggests:

1. MCP Gateway starts but encounters error during initialization
2. Tavily MCP server may be configured but not connecting properly
3. Issue may be different from Daily News (which recovered after TAVILY_API_KEY)

Recommended Next Steps

1. Download and analyze artifacts from run 21304877267

   • Check /tmp/gh-aw/mcp-logs/ for detailed error messages
   • Review agent stdio logs for MCP Gateway startup sequence

2. Compare MCP Gateway configuration with Daily News workflow

   • Daily News: ✅ Working (TAVILY_API_KEY fixed it)
   • MCP Inspector: ❌ Failing (TAVILY_API_KEY exists but still fails)
   • Difference: May need additional configuration or different MCP server setup

3. Test manually with increased logging

   • Trigger workflow_dispatch with debug mode
   • Check MCP Gateway version compatibility
   • Verify Docker container accessibility

Priority: P1 (High)

This is blocking MCP tooling inspection for 19 days. Urgent investigation needed.

AI generated by Workflow Health Manager - Meta-Orchestrator

[github-actions]

This issue was automatically closed because it expired on 2026-01-24T02:59:47.003Z.