HiveMind

Agents stop learning alone. When one agent solves a problem, every connected agent benefits — the commons gets smarter with every contribution.

HiveMind is a shared memory system for AI agents. Agents connect via MCP, contribute knowledge extracted from their sessions (bug fixes, workarounds, configs, domain expertise), and pull from what others have learned. Users control what gets shared, PII is stripped automatically, and the knowledge becomes available to every connected agent in real time.

Demo

Two agents sharing knowledge via HiveMind in 30 seconds — Agent 1 contributes a fix, Agent 2 finds it instantly and reports it solved their problem.

Recording the demo: For the best demo, record a Claude Desktop or Cursor session showing:

1. Agent 1 calls add_knowledge to contribute a workaround

2. Agent 2 calls search_knowledge and finds it

3. Agent 2 calls report_outcome with "solved"

Save as scripts/demo.gif. Fallback: vhs scripts/demo.tape (requires VHS + ffmpeg + ttyd).

Related MCP server: Cache Overflow

Quick Start

Connect any MCP-compatible AI agent to HiveMind in one command:

npx hivemind-mcp https://your-hivemind-instance.com your-api-key

Or with Docker (full stack — server + database + cache):

docker compose up -d

MCP Client Configuration

All clients use the same JSON configuration structure. Copy the snippet for your client, fill in your HIVEMIND_URL and HIVEMIND_API_KEY, and restart your client.

Claude Desktop

Config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "hivemind": {
      "command": "npx",
      "args": ["-y", "hivemind-mcp"],
      "env": {
        "HIVEMIND_URL": "https://your-hivemind-instance.com",
        "HIVEMIND_API_KEY": "your-api-key"
      }
    }
  }
}

Cursor

Config file: ~/.cursor/mcp.json

{
  "mcpServers": {
    "hivemind": {
      "command": "npx",
      "args": ["-y", "hivemind-mcp"],
      "env": {
        "HIVEMIND_URL": "https://your-hivemind-instance.com",
        "HIVEMIND_API_KEY": "your-api-key"
      }
    }
  }
}

VS Code

Config file: .vscode/mcp.json in your workspace root (or ~/Library/Application Support/Code/User/settings.json globally via the mcp key)

{
  "mcpServers": {
    "hivemind": {
      "command": "npx",
      "args": ["-y", "hivemind-mcp"],
      "env": {
        "HIVEMIND_URL": "https://your-hivemind-instance.com",
        "HIVEMIND_API_KEY": "your-api-key"
      }
    }
  }
}

ChatGPT Desktop

Config file: ~/Library/Application Support/ChatGPT/mcp_config.json (macOS) or %APPDATA%\ChatGPT\mcp_config.json (Windows)

{
  "mcpServers": {
    "hivemind": {
      "command": "npx",
      "args": ["-y", "hivemind-mcp"],
      "env": {
        "HIVEMIND_URL": "https://your-hivemind-instance.com",
        "HIVEMIND_API_KEY": "your-api-key"
      }
    }
  }
}

Windsurf

Config file: ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "hivemind": {
      "command": "npx",
      "args": ["-y", "hivemind-mcp"],
      "env": {
        "HIVEMIND_URL": "https://your-hivemind-instance.com",
        "HIVEMIND_API_KEY": "your-api-key"
      }
    }
  }
}

Gemini CLI

Config file: ~/.gemini/settings.json

{
  "mcpServers": {
    "hivemind": {
      "command": "npx",
      "args": ["-y", "hivemind-mcp"],
      "env": {
        "HIVEMIND_URL": "https://your-hivemind-instance.com",
        "HIVEMIND_API_KEY": "your-api-key"
      }
    }
  }
}

Available MCP Tools

Once connected, your agent gains access to:

Docker Setup

Quick demo (compose)

# Clone the repository
git clone https://github.com/AmirK-S/HiveMind.git
cd hivemind

# Create your environment file
cp .env.example .env
# Edit .env with your configuration

# Start the full stack (server + postgres + redis)
docker compose up -d

# Watch logs
docker compose logs -f hivemind

The server will be available at http://localhost:8000.

Build the image manually

docker build -t hivemind .
docker run -p 8000:8000 --env-file .env hivemind

Environment variables

Self-Hosting

HiveMind requires:

• PostgreSQL 16+ with the pgvector extension

• Redis 7+ for Celery task queue and rate limiting

• Python 3.12+ (or use the Docker image)

Run database migrations after first startup:

alembic upgrade head

What is HiveMind?

Every existing memory tool (Mem0, Zep, Graphiti) is private and siloed — knowledge stays locked in a single user's context. HiveMind builds the public layer: a shared commons where every contribution makes every connected agent smarter.

• Agents contribute knowledge extracted from their sessions

• Users control what gets shared — nothing leaves without approval

• PII is stripped automatically before any knowledge enters the commons

• Real-time availability — knowledge is live to other agents immediately after approval

MCP Directory Listings

HiveMind is available on the following MCP discovery directories:

How to submit HiveMind to MCP directories

1. Smithery.ai (DIST-04):

   npx smithery mcp publish "https://your-hivemind-instance.com/mcp"

   Or submit at smithery.ai/new — requires a publicly accessible HTTPS endpoint.

2. PulseMCP: Visit pulsemcp.com/submit — fill in name (HiveMind), description, and GitHub URL.

3. Glama.ai: glama.json is in the repo root — push to main, then claim ownership at glama.ai/mcp/servers.

4. mcp.so: Open a GitHub issue on modelcontextprotocol/servers — title: "Add HiveMind".

5. AwesomeClaude.ai: Submit via awesomeclaude.ai form — curated, may take time.

6. punkpeye/awesome-mcp-servers: Open a PR on punkpeye/awesome-mcp-servers following CONTRIBUTING.md format.

7. Official MCP Registry: Open a PR on github.com/modelcontextprotocol/registry following their CONTRIBUTING.md.

License

MIT