ARCHIVED 2026-03-23 — Superseded by ultimate-claude-code-skills-playbook.md. Archived copy at 2026-03.

Solanasis MCP & Plugins Installation — Quick Checklist

Date: March 15, 2026 For: Dmitri Zasage Status: Ready to execute Time Estimate: Phase 1 (4-6 hours) + Phase 2 (2-3 hours) + Phase 3 (1 week monitoring)

CRITICAL: Do Phase 1-2 on your local machine, NOT in Cowork. Cowork is the VM — installations don’t persist.

PHASE 1: MCP SERVERS & CONFIGURATION (LOCAL MACHINE)

[ ] Step 1: Install DuckDuckGo MCP

npm install -g duckduckgo-mcp
which duckduckgo-mcp  # Verify

[ ] Step 2: Configure DuckDuckGo in ~/.claude/settings.json

{
  "mcpServers": {
    "duckduckgo": {
      "command": "npx",
      "args": ["-y", "@duckduckgo/duckduckgo-mcp"]
    }
  }
}

[ ] Step 3: Add Exa MCP (remote endpoint)

{
  "mcpServers": {
    "exa": {
      "command": "curl",
      "args": ["-sL", "https://mcp.exa.ai/mcp?tools=web_search_exa,web_search_advanced_exa,company_research_exa,people_search_exa,get_code_context_exa,crawling_exa,deep_researcher_start,deep_researcher_check"]
    }
  }
}

[ ] Step 4: Disable Claude.ai MCP inheritance (CRITICAL)

# At terminal, run Claude with this flag:
ENABLE_CLAUDEAI_MCP_SERVERS=false claude
 
# Or add to ~/.bashrc or ~/.zshrc:
export ENABLE_CLAUDEAI_MCP_SERVERS=false

[ ] Step 5: Verify MCP Tools

claude
/mcp
# Should see: duckduckgo, exa

PHASE 1B: CLAUDE.md FILES (LOCAL)

[ ] Step 6: Create ~/.claude/CLAUDE.md

File path: ~/.claude/CLAUDE.md
Source: See full content in INSTALLATION-PLAN document (Section 1B, Step 5)
Command: cat > ~/.claude/CLAUDE.md << 'EOF' + paste content + EOF

[ ] Step 7: Update /solanasis-docs/.claude/CLAUDE.md

Read existing file first (don’t overwrite security rules)
Add MCP sections from INSTALLATION-PLAN document
Keep all existing content

PHASE 1C: SUBAGENT FILES (LOCAL)

[ ] Step 8: Create ~/.claude/agents/ directory

mkdir -p ~/.claude/agents

[ ] Step 9: Create senior-reviewer.md

File path: ~/.claude/agents/senior-reviewer.md
Source: INSTALLATION-PLAN document (Section 1C, Step 8)
Verify: cat ~/.claude/agents/senior-reviewer.md | head -10

[ ] Step 10: Create planner.md

File path: ~/.claude/agents/planner.md
Source: INSTALLATION-PLAN document (Section 1C, Step 9)
Verify: cat ~/.claude/agents/planner.md | head -10

[ ] Step 11: Create research-agent.md

File path: ~/.claude/agents/research-agent.md
Source: INSTALLATION-PLAN document (Section 1C, Step 10)
Verify: cat ~/.claude/agents/research-agent.md | head -10

[ ] Step 12: Test subagents in Claude

claude
# Test: "Plan a task to research three prospects"
# Test: "Research the compliance landscape for SMBs"
# Test: "Create a client proposal, then ask for review"

PHASE 2: MARKETPLACE PLUGINS (LOCAL CLI)

[ ] Step 13: Install 6 Phase 1 plugins

claude
 
# Run in CLI:
/plugin install operations@knowledge-work-plugins
/plugin install sales@knowledge-work-plugins
/plugin install marketing@knowledge-work-plugins
/plugin install engineering@knowledge-work-plugins
/plugin install customer-support@knowledge-work-plugins
/plugin install data@knowledge-work-plugins
 
# Verify:
/plugin list

[ ] Step 14: Disable model invocation for 8 skills (CONTEXT BUDGET)

Set these 8 skills to MANUAL-ONLY (lookup in plugin config files or CLI):

change-management
resource-planning
create-an-asset
competitive-intelligence
daily-briefing
campaign-planning
knowledge-management
data-validation

Keep these 12 AUTO-INVOKE:

compliance-tracking
risk-assessment
vendor-management
process-optimization
account-research
draft-outreach
call-prep
content-creation
documentation
incident-response
data-exploration
sql-queries

PHASE 2B: CUSTOM SKILLS

[ ] Step 15: Create research-first skill

Path: ~/.claude/skills/research-first/SKILL.md
Source: INSTALLATION-PLAN document (Section 2B, Step 15)
Test: /research-first [topic]

PHASE 3: VERIFICATION (LOCAL)

[ ] Step 16: Run All Tests

Test	Command	Expected Result
MCP Tools	`claude` → `/mcp`	Output: duckduckgo, exa
DuckDuckGo	”Search for SOC 2 compliance”	Uses DuckDuckGo tool, not Chrome
Exa	”Research Anthropic Inc”	Uses Exa tools, returns company data
Chrome for Interactive	”Log into ClickUp and show sprint”	Uses Chrome (login required)
Planner	”Plan a task to research prospects”	planner subagent invokes
Research Agent	”Deep research SMB compliance”	research-agent subagent invokes
Senior Reviewer	”Create a proposal, then review”	senior-reviewer subagent invokes
Plugin Skills	”What skills do I have?”	Lists 20 skills (12 auto + 8 manual)
Context Budget	”Any ‘Excluded skills’ warnings?"	"No” = context budget OK

All tests should pass. If any fail, see TROUBLESHOOTING in INSTALLATION-PLAN document.

PHASE 3: COWORK SESSION TEST

[ ] Step 17: Open Cowork and test end-to-end

Open Cowork with solanasis-docs folder
Check that plugins appear in skill menu
Run full workflow:
- Research a prospect (uses research-agent + Exa/WebSearch)
- Draft outreach (uses sales skills)
- Create asset (uses create-an-asset skill)
- Ask for review (uses senior-reviewer)

FIRST WEEK MONITORING

[ ] Day 1: Verify all connections

Test each MCP tool
Test each subagent
Test 5 critical plugins

[ ] Day 2-3: Run real workflows

Prospect research → outreach → asset creation
Security assessment workflow
Data migration workflow

[ ] Day 4-5: Monitor Chrome usage

Track how often Chrome is used for research
Goal: Chrome < 20% for research-only tasks
If > 20%: Consider adding PreToolUse hook (optional)

[ ] Day 6-7: Adjust model invocation

Identify which manual-only skills should auto-invoke
Re-enable any skills used > 3x/day
Disable any skills that auto-fire incorrectly

[ ] End of Week: Measure ROI

Count deliverables using plugins
Count issues found by senior-reviewer
Count research tasks using research-agent
Decision: Keep all or disable low-value plugins?

TROUBLESHOOTING QUICK REFERENCE

MCP not appearing after install:

# Restart Claude
exit
claude
/mcp

Context overflow (Excluded skills warning):

Disable more skills to manual-only
Or remove plugins temporarily
See context budget section in INSTALLATION-PLAN

Subagents not auto-invoking:

Check YAML frontmatter (no syntax errors)
Verify description field includes “use proactively”
Restart Claude: exit then claude

Chrome being used for research:

Verify CLAUDE.md instructions are loaded: cat ~/.claude/CLAUDE.md | head -5
Test again with explicit instruction: “Search for [topic] using WebSearch”
After week 1: Consider adding PreToolUse Chrome gate hook

Plugin installation fails:

Try: /plugin marketplace search [plugin-name]
Then click install from results
If still fails: Check Claude Code version (claude --version)

SUCCESS CRITERIA

Installation is done when:

✅ MCP tools (Exa, DuckDuckGo) available and tested ✅ CLAUDE.md files created (user + project level) ✅ All 3 subagents created and auto-invoke correctly ✅ All 6 Phase 1 plugins installed (12 auto, 8 manual) ✅ Custom skill (research-first) created ✅ Context budget OK (no “Excluded skills” warnings) ✅ All 8 verification tests pass ✅ Full end-to-end workflow works in Cowork

NEXT STEPS AFTER INSTALLATION

Week 1: Monitor Chrome usage, adjust model invocation mapping
Week 2: Review effectiveness, decide on Phase 2 plugins (GitHub skills)
Month 2: Build custom skills (security assessment, DR verification, data migration planner)
Month 2+: Evaluate Wyre MSP platform, HR plugin, advanced features

See Master Setup Plan for full roadmap.

REFERENCE DOCUMENTS

INSTALLATION-PLAN-MCP-PLUGINS-SKILLS.md — Complete step-by-step guide (this is the detailed version)
solanasis-cowork-master-setup-plan.md — Overall strategy and roadmap
solanasis-claudemd-and-agent-architecture.md — CLAUDE.md design, subagent details, verification strategies
claude-skills-acquisition-plan.md — Evaluation of every plugin and skill

Ready to go. Dmitri, review this plan and let me know when you’re ready to start Phase 1.

Solanasis Docs

Explorer

INSTALLATION-QUICK-CHECKLIST