Installation

Quoth can be connected to any MCP-compatible AI tool. This guide covers installation for popular platforms.

Claude Code (CLI)

Local Plugin (Recommended)

The Quoth local plugin provides autonomous self-learning with 22 MCP tools, 9 hook bindings, and a background daemon.

Interactive Setup (Recommended)

# Clone the repo
git clone https://github.com/Montinou/quoth.git
cd quoth

# Interactive CLI wizard
node quoth-plugin/scripts/cli.js init

The wizard will prompt you to choose between two modes:

Manual Setup

# Run setup script (idempotent — safe to re-run)
bash quoth-plugin/scripts/setup.sh

What You Get

• 22 local MCP tools: 8 patterns, 6 intelligence, 6 agents, 2 skills
• 9 hooks: trajectory capture, task routing, pattern injection, Bayesian feedback
• Background daemon: JUDGE → DISTILL → CONSOLIDATE pipeline + session batch distill
• Unix socket query server: ~/.quoth/daemon.sock for ~20ms routing and pattern injection
• Batch embeddings: generateEmbeddingBatch() for cost-optimized vector generation
• Context enrichment: user intent, LLM reasoning, sanitized I/O in every trajectory
• Autonomous doc maintenance: nightly content-hash-based doc updates with git push

Embedding Provider

The daemon uses Venice BGE-M3 for cloud embeddings (migrated from voyage-4-lite in v3.3.0). For local mode, configure your embedding API key:

# Venice API key (for local embeddings via Venice BGE-M3)
export VENICE_API_KEY="your-key"

Daemon Socket

The daemon exposes a Unix socket at ~/.quoth/daemon.sock for fast IPC. Hooks use this socket for ~20ms routing and pattern injection instead of spawning subprocesses.

# Check daemon status (includes socket health)
quoth_daemon_status

Cloud MCP (Cross-Team Features)

For cross-team knowledge sharing, semantic search, and agent-to-agent communication:

claude mcp add --transport http quoth https://quoth.triqual.dev/api/mcp

Then authenticate from the /mcp menu.

Token-Based Installation

claude mcp add --transport http quoth https://quoth.triqual.dev/api/mcp \
  --header "Authorization: Bearer YOUR_TOKEN"

Both Together

Install both for the full experience. Local plugin handles self-learning, cloud MCP handles team sharing. High-confidence patterns auto-promote nightly.

Claude Desktop

Add to your Claude Desktop configuration file:

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

Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "quoth": {
      "url": "https://quoth.triqual.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

Cursor

Cursor supports MCP through its settings:

SSE Transport

For clients that require Server-Sent Events transport:

claude mcp add --transport sse quoth \
  "https://quoth.triqual.dev/api/sse?token=YOUR_TOKEN"

Agent-to-Agent Bus Setup

To enable cross-instance agent communication, register your agent after installation:

// Get your organization agents
quoth_agent_list()

// Register this agent instance
quoth_agent_register({
  agent_name: "my-local-agent",
  display_name: "My Local Agent 🖥️",
  instance: "mac",  // or "aws", "montino", etc.
  role: "specialist"
})

Once registered, you can:

• Send messages to agents on other instances (with HMAC signing)
• Create tasks for other agents
• Receive real-time notifications via async message bus

Verify Installation

Test that Quoth is connected:

Search Quoth for "test patterns"

If connected, you'll see search results from the Quoth knowledge base.

Test the A2A Bus:

// List agents in your organization
quoth_agent_list()

Troubleshooting

Common Issues

Agent Communication Issues

If agent-to-agent messaging isn't working:

1. Verify both agents are registered and active: quoth_agent_list()
2. Check the recipient's inbox: quoth_agent_inbox({ agent_name: "recipient" })
3. Ensure the receiving agent is listening for realtime events
4. Verify HMAC signatures are being generated (automatic)