hivemind

Instant troubleshooting solutions for AI coding assistants Community-driven knowledge base with 16,000+ solutions and growing

Works with any MCP-compatible CLI: Claude Code, Codex CLI, Gemini CLI, Grok CLI, Cursor, and more.

🎯 What is hivemind?

hivemind is a Model Context Protocol (MCP) server that gives AI coding assistants instant access to a searchable knowledge base of troubleshooting solutions.

Supported platforms:

• Claude Code (Anthropic)

• Codex CLI (OpenAI)

• Gemini CLI (Google)

• Grok CLI (xAI)

• Cursor

• Any MCP-compatible tool

When you hit an error, search hivemind - get ranked solutions from the community in seconds.

✨ Features

• 🔍 Full-text search - Fast Postgres FTS with relevance ranking

• 👍 Community feedback - Say "hivemind: worked" to rate solutions

• 🎯 Trigger words - AI automatically tracks your feedback

• 📊 Smart ranking - Solutions sorted by success rate and votes

• 🚀 Zero config - No API keys, no setup, just install and use

• ⚡ Fast - < 1s search response time

📦 Installation

Quick Install

Claude Code:

Codex CLI:

Gemini CLI:

Cursor / Windsurf:

Manual Setup (Claude Code)

Option A: Via Claude Code UI

1. Open Claude Code settings

2. Go to MCP Servers section

3. Add new server:

   • Name: hivemind

   • Command: npx

   • Args: ["hivemind-mcp"]

Option B: Via config file

Edit ~/.config/claude/mcp_config.json:

Restart your CLI/editor to activate the MCP server.

🚀 Usage

Search for Solutions

Or explicitly:

Give Feedback

After trying a solution:

Or:

Your AI automatically tracks your feedback to improve rankings!

📚 Example Queries

Current knowledge base: 16,000+ solutions (and growing) covering MCP, Playwright, Supabase, web development, databases, security, and more.

🤝 Contributing Solutions

Have a solution that helped you? Share it with the community!

Via MCP Tool

Via Email

Send to: [your-email] (Beta testers will get this in welcome message)

Include:

• Problem/error message

• Category (mcp-troubleshooting, web-automation, etc.)

• Step-by-step solutions

• What to check before/after

• Common mistakes

🎓 How Trigger Words Work

When you say "hivemind: worked" or "hivemind: failed":

1. Your AI recognizes the trigger phrase

2. Automatically calls feedback tool

3. Backend increments thumbs_up or thumbs_down

4. Future searches show better-ranked solutions

Why trigger words? Natural language ("that worked") is ambiguous. Trigger phrases ensure your AI ALWAYS catches your feedback.

🛡️ Infrastructure

Rate Limits:

• Search: 100 requests/hour per IP

• Voting: 20 votes/hour per IP

• Contributions: 5/hour per IP

Stack:

• Backend: Supabase (Postgres 17 + Edge Functions)

• Database: 16,000+ solutions

• Security: IP banning, input sanitization, moderated contributions

🔒 Security

hivemind is designed to be a safe, public knowledge base. We take security seriously:

Credential Scanning

All contributions are scanned for sensitive data before storage. Our scanner blocks:

• API Keys: OpenAI, Anthropic, AWS, GCP, Azure, Stripe, Slack, GitHub, GitLab, and 30+ providers

• Tokens: JWT, Bearer tokens, OAuth tokens, PATs

• Secrets: Database passwords, webhook URLs, Basic auth headers

• Obfuscation Attempts: Base64-encoded secrets, zero-width character tricks

How it works:

1. Pattern matching against 45+ known secret formats

2. Base64 decode + rescan for hidden secrets

3. Entropy analysis for high-randomness strings near keywords like key=, secret=

4. Zero-width character stripping to defeat obfuscation

If your contribution is rejected, check for accidentally included credentials.

Row-Level Security

All database tables use Postgres RLS policies. The public API can only:

• Read solutions and skills

• Submit contributions (to moderation queue)

• Submit feedback votes

No direct write access to production data.

Rate Limiting

Aggressive rate limits prevent abuse:

• Search: 100/hour

• Voting: 20/hour

• Contributions: 5/hour

What We Don't Store

• No user accounts required

• No tracking cookies

• No personally identifiable information

• IP addresses used only for rate limiting (not logged with searches)

See PRIVACY.md for full privacy policy.

📊 Monitoring

Beta testers: We monitor usage via Supabase dashboard. You can check your own usage:

See MONITORING.md for admin dashboard access.

🗺️ Roadmap

Completed:

• ✅ Core search and feedback

• ✅ Trigger word system

• ✅ Rate limiting

• ✅ Security (IP bans, sanitization, moderation queue)

• ✅ 16,000+ solutions

• ✅ Version update notifications

Coming Soon:

• Semantic search (AI embeddings)

• Contributor leaderboard

• Browser extension

• Dangerous command blocklist

See ROADMAP.md for full details.

🔧 Technical Details

Stack:

• MCP Server: TypeScript + @modelcontextprotocol/sdk

• Backend: Supabase (Postgres 17 + Deno Edge Functions)

• Search: Postgres Full-Text Search (ts_rank)

• Hosting: Supabase free tier + npm registry

Architecture:

Public Source:

• hivemind-mcp/ - MCP client package (what you install via npm)

📄 Documentation

• INSTALL.md - Detailed installation guide

• TRIGGER_FEEDBACK_SYSTEM.md - How feedback works

• MONITORING.md - Admin monitoring guide

• ROADMAP.md - Product roadmap

❓ Troubleshooting

MCP not loading

Search not working

1. Check your AI CLI MCP logs

2. Verify internet connection

3. Try: "Search hivemind for test"

Feedback not tracked

Known issue - tracking endpoint works but MCP may not call it consistently. We're debugging this.

🐛 Reporting Issues

1. Check existing issues

2. Create new issue with:

   • What you tried

   • What happened

   • Your AI CLI (Claude Code, Codex, Gemini, etc.)

   • Operating system

📜 License

MIT License - see LICENSE file

🌟 Star History

If hivemind helped you, give it a star ⭐

Built with ❤️ for the AI coding community