hugo-memex

MCP server that makes any Hugo site's content queryable via SQLite with FTS5 full-text search.

Your Hugo blog is ground truth. hugo-memex indexes it into a SQLite cache with full-text search, taxonomy queries, and JSON front matter extraction : then exposes it via MCP so AI assistants can query, search, and create content.

Install

pip install hugo-memex

Quick Start

# Configure
mkdir -p ~/.config/hugo-memex
cat > ~/.config/hugo-memex/config.yaml << 'EOF'
hugo_root: ~/path/to/your/hugo-site
database_path: ~/.config/hugo-memex/hugo.db
EOF

# Index your site
hugo-memex index

# Query it
hugo-memex search "machine learning"
hugo-memex sql "SELECT title, section, date FROM pages ORDER BY date DESC LIMIT 10"
hugo-memex stats

MCP Integration

Add to your .mcp.json for Claude Code:

{
  "mcpServers": {
    "hugo-memex": {
      "command": "python",
      "args": ["-m", "hugo_memex", "mcp"],
      "env": {
        "HUGO_MEMEX_CONFIG": "/path/to/config.yaml"
      }
    }
  }
}

Tools

Resources

Architecture

• DB is a read-only cache : Hugo content files are ground truth
• Generic schema : JSON front_matter column, no per-content-type tables
• Taxonomies auto-discovered from hugo.toml
• Incremental sync via SHA-256 content hash + file mtime
• FTS5 with porter stemming + unicode61 tokenizer
• SQLite authorizer enforces read-only (not bypassable via PRAGMA)
• Raw sqlite3 : no ORM. WAL mode, foreign keys.

Configuration

# ~/.config/hugo-memex/config.yaml
hugo_root: ~/github/repos/my-hugo-site   # contains hugo.toml + content/
database_path: ~/.config/hugo-memex/hugo.db

Environment variable overrides: HUGO_MEMEX_CONFIG, HUGO_MEMEX_HUGO_ROOT, HUGO_MEMEX_DATABASE_PATH.

License

MIT