Honeypot MCP Server

A small Model Context Protocol server that exposes my live honeypot's threat-intelligence database to an MCP client (Claude Desktop / Claude Code) as read-only tools — so I can investigate attacker activity by just asking, instead of writing SQL.

The data comes from a self-hosted honeypot stack (Cowrie SSH/Telnet + a custom HTTP honeypot) writing into PostgreSQL — currently ~530k attack sessions, ~33k login attempts, and thousands of captured attacker commands and file-staging events.

Tools

Related MCP server: FastMCP ThreatIntel

Safety

• Read-only by construction. Every connection opens a read-only transaction and every query is a SELECT. No tool mutates data.

• The IP passed to lookup_ip is validated with ipaddress and bound as a query parameter — never string-formatted into SQL.

• Recommended: point HONEYPOT_DATABASE_URL at a DB role granted SELECT only.

Setup

python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env          # then edit with your connection string
export HONEYPOT_DATABASE_URL="postgresql://user:pass@127.0.0.1:5433/honeypot"

python server.py selftest      # verify DB connectivity
python server.py               # run as an MCP (stdio) server

Connecting a client

Claude Code (claude mcp add):

claude mcp add honeypot -- bash -lc 'cd /path/to/honeypot-mcp && \
  HONEYPOT_DATABASE_URL="postgresql://user:pass@127.0.0.1:5433/honeypot" \
  .venv/bin/python server.py'

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "honeypot": {
      "command": "/path/to/honeypot-mcp/.venv/bin/python",
      "args": ["/path/to/honeypot-mcp/server.py"],
      "env": { "HONEYPOT_DATABASE_URL": "postgresql://user:pass@127.0.0.1:5433/honeypot" }
    }
  }
}

The DB lives on my server (bound to localhost), so I either run this server there, or launch it over SSH stdio from my laptop:

{ "mcpServers": { "honeypot": {
  "command": "ssh",
  "args": ["ubuntu", "cd honeypot-mcp && .venv/bin/python server.py"]
}}}