Fix MCP Inspector timeout by removing unnecessary server imports

[Copilot]

Problem

MCP Inspector workflow failing for 20 days with 0% success rate. "Start MCP gateway" step times out after 2 minutes trying to initialize 17 MCP servers, several with missing credentials (DD_API_KEY, CONTEXT7_API_KEY, etc.) causing 401 errors.

Root Cause

Workflow purpose: READ and DOCUMENT MCP config files in .github/workflows/shared/mcp/*.md.

Actual behavior: Importing and initializing all 15 MCP servers (arxiv, ast-grep, brave, context7, datadog, deepwiki, fabric-rti, gh-aw, markitdown, microsoft-docs, notion, sentry, server-memory, slack, tavily) unnecessarily.

Changes

Removed unnecessary MCP server imports

 imports:
-  - shared/mcp/arxiv.md
-  - shared/mcp/ast-grep.md
-  # ... 13 more MCP servers
-  - shared/mcp/tavily.md
+  # MCP Inspector only needs to READ the MCP config files, not USE the servers
   - shared/reporting.md

Impact:

• MCP servers: 17 → 3 (github, safeoutputs, serena)
• Lock file: -780 lines
• Docker images: 9 → 3
• Gateway startup: 2+ min timeout → expected <10s
• Removed: Node.js, Python, uv, Go setup steps

Workflow retains all tools needed for its actual purpose: edit, bash, serena (Go navigation), cache-memory.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>MCP Inspector - Still failing after TAVILY_API_KEY fix (0% success rate, 20 days offline)</issue_title>
<issue_description>### Problem

MCP Inspector workflow remains non-operational with 0% success rate despite TAVILY_API_KEY secret being added. The workflow has been offline for 20 days.

Current Status (2026-01-25)

• Success rate: 0/5 recent runs (0%)
• Latest failure: §21304877267 (2026-01-23)
• Days offline: 20 days (since 2026-01-05)
• Failed step: "Start MCP gateway" (step 24)
• Previous issue: Fix MCP Inspector workflow - "Start MCP gateway" failure (80% failure rate) #11433 (auto-closed 2026-01-24)

Root Cause Analysis

Key insight: Daily News workflow recovered immediately after TAVILY_API_KEY was added (2026-01-22), but MCP Inspector did NOT recover. This suggests:

1. Hypothesis 1: Workflow needs recompilation

   • Secret was added AFTER last compilation
   • Lock file may not reference the new secret
   • Solution: make recompile

2. Hypothesis 2: Different MCP Gateway configuration

   • MCP Inspector may use different MCP server setup than Daily News
   • May need additional configuration beyond TAVILY_API_KEY
   • Review frontmatter differences

3. Hypothesis 3: MCP Gateway version incompatibility

   • Check Docker container versions
   • Verify MCP Gateway version matches requirements

Comparison with Daily News

Aspect	Daily News (✅ Working)	MCP Inspector (❌ Failing)
TAVILY_API_KEY	Present	Present
Recovery	Immediate (2026-01-22)	None
Success rate	40% (recovering)	0% (failing)
Last compiled	Unknown	Unknown
MCP Gateway	Working	Failing at startup

Recommended Investigation Steps

Step 1: Recompile Workflow

cd /path/to/repo
make recompile
git add .github/workflows/mcp-inspector.lock.yml
git commit -m "Recompile MCP Inspector after TAVILY_API_KEY fix"
git push

Step 2: Compare Frontmatter

Compare configurations:

• .github/workflows/daily-news.md (working)
• .github/workflows/mcp-inspector.md (failing)

Look for differences in:

• MCP server configuration
• Tool permissions
• Environment variables

Step 3: Analyze Failed Run Logs

Download artifacts from run 21304877267:

• Check /tmp/gh-aw/mcp-logs/ for MCP Gateway errors
• Review agent stdio logs
• Look for secret access issues

Step 4: Test Manually

gh workflow run mcp-inspector.lock.yml

Monitor run and check logs immediately.

Success Criteria

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

Priority: P1 (High)

Impact: Critical tooling for MCP debugging offline for 20 days. This blocks MCP server development and troubleshooting.

Urgency: High - affecting developer productivity and MCP ecosystem health.

Next steps:

1. Recompile workflow (5 min)
2. Test manually (10 min)
3. Analyze results and iterate (30 min)

References:

• Latest failed run §21304877267
• Workflow source
• Previous issue githubnext/gh-aw#11433 (auto-closed)
• Dashboard githubnext/gh-aw#11581

AI generated by Workflow Health Manager - Meta-Orchestrator

• expires on Jan 26, 2026, 3:08 AM UTC

Comments on the Issue (you are @copilot in this section)

• Fixes MCP Inspector - Still failing after TAVILY_API_KEY fix (0% success rate, 20 days offline) #11721

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[pelikhan]

@copilot MCP-inspector should start servers concurrently

[Copilot]

@copilot MCP-inspector should start servers concurrently

The MCP gateway currently starts servers sequentially, and there's no configuration option in the workflow frontmatter or MCP gateway config schema to enable concurrent startup. This would require changes to the MCP gateway container itself (ghcr.io/githubnext/gh-aw-mcpg:v0.0.78).

The current fix in this PR removes the unnecessary MCP server imports (reducing from 17 to 3 servers), which eliminates the timeout issue for this specific workflow. However, if you'd like concurrent server startup as a general improvement, that would need to be implemented in the MCP gateway codebase and then a new configuration option added to the schema in pkg/workflow/schemas/mcp-gateway-config.schema.json (perhaps "concurrentStartup": true in the gateway config section).