meshpop-db 1.0.2

Distributed search database with semantic indexing and MCP server

MeshDB

Distributed full-text search engine with LLM conversation indexing.

Part of MeshPOP — Layer 4 (Search)

Features

• Full-text search with SQLite FTS5 across all servers (BM25 ranking, snippets)
• LLM Conversation Indexing — indexes Q&A from Claude Code, Cowork, Cursor, ChatGPT, Gemini across all 14 servers
• Multi-server discovery — automatically finds and indexes conversations on remote servers via vssh protocol
• MCP-driven discover — connected AIs can register their own conversation storage paths
• Semantic search with vector embeddings (ChromaDB)
• Auto-indexing with file change detection (hash-based skip)

Install

pip install meshpop-db

Usage

Document Search

meshdb search "nginx config"         # Full-text search across all servers
meshdb find "config.json"            # Find files by name
meshdb semantic "how to deploy"      # Semantic vector search
meshdb status                        # Database status

Conversation Search

# Index local conversations (Claude Code, Cursor, ChatGPT, Gemini)
meshdb-conv index

# Index local + all remote servers (d1, d2, g1, g2, g3)
meshdb-conv index --remote

# Search across all indexed AI conversations
meshdb-conv search "wireguard setup"
meshdb-conv search "PyPI upload" --source claude --project MeshPOP

# Show what conversation files exist on this machine
meshdb-conv discover

# Status — shows chunks by source, server, project
meshdb-conv status

JSON Agent Protocol (for remote vssh calls)

meshdb-conv json discover           # List discovered session files as JSON
meshdb-conv json chunks <filepath>  # Parse file into Q&A chunks as JSON
meshdb-conv json search <query>     # Search and return JSON
meshdb-conv json status             # Stats as JSON

MCP Tools

Add to Claude Desktop / Claude Code config:

{
  "mcpServers": {
    "meshdb": {
      "command": "python3",
      "args": ["/path/to/meshdb-mcp-server.py"]
    }
  }
}

Available tools:

Tool	Description
meshdb_search	Full-text search across all servers (244K+ files)
meshdb_find	Find files by filename
meshdb_read	Read indexed file content
meshdb_conversations	Search LLM conversations (Claude, Cursor, ChatGPT, Gemini)
meshdb_conv_read	Read full Q&A chunk
meshdb_conv_discover	Discover/register conversation sources; trigger indexing
meshdb_semantic	Semantic code search via ChromaDB
meshdb_status	Database status across all servers

Architecture

m1 (macOS) ─── direct SQLite ─── meshdb.db (central)
  │
  ├── vssh ──→ d2: json discover → json chunks → insert
  ├── vssh ──→ g1: json discover → json chunks → insert
  ├── vssh ──→ g2: json discover → json chunks → insert
  ├── vssh ──→ g3: json discover → json chunks → insert
  └── vssh ──→ d1: json discover → json chunks → insert

Conversation sources discovered automatically:

Source	Format	Path
Claude Code	JSONL	~/.claude/projects/**/*.jsonl
Cowork	JSONL	~/Library/.../local-agent-mode-sessions/**/*.jsonl
Cursor	SQLite	~/.config/Cursor/User/workspaceStorage/*/state.vscdb
ChatGPT	JSON	~/Downloads/chatgpt_export/conversations.json
Gemini	JSONL/JSON	~/.gemini/history/**/*

MCP-driven discovery: Connected AIs can register new sources via meshdb_conv_discover tool — no code changes needed.

Current Stats

• 34,500+ Q&A chunks indexed across 7 servers
• 1,589 unique sessions from Claude Code, Cowork, Cursor
• Coverage: 2025-12 to 2026-03
• Indexing speed: ~17s for 231 remote files across 5 servers

Links

• Main project: github.com/meshpop/mpop
• Website: mpop.dev
• PyPI: pypi.org/project/meshpop-db

License

Apache-2.0