Introducing the Quoth Plugin for Claude Code

We've rebuilt how Quoth integrates with Claude Code. Instead of heavy persona prompts that cost thousands of tokens, the new Quoth Plugin uses lightweight hooks and an adaptive tool that Claude calls when it needs guidance.

The Problem with Personas

Our original approach used three MCP prompts—Architect, Auditor, and Documenter—each weighing in at ~2,000 tokens. Combined with verbose hooks, a typical session consumed ~750 tokens before you even asked a question.

For power users running dozens of sessions daily, this added up fast.

The New Approach: Hints + Autonomy

The Quoth Plugin flips the model:

How It Works

1. Session Start: A Gentle Hint

When you start a Claude Code session with Quoth connected, you'll see:

Quoth MCP active. Strongly recommend `quoth_guidelines('code')`
and `quoth_search_index` before writing code.

That's it. ~25 tokens. Claude now knows Quoth is available.

2. The quoth_guidelines Tool

Instead of three separate prompts, there's now one adaptive tool:

quoth_guidelines({
  mode: "code" | "review" | "document",
  full?: boolean  // compact by default
})

Compact mode (~150 tokens) gives Claude the core rules:

• Search first with quoth_search_index
• Trust levels: >80% follow exactly, 60-80% verify, <60% cross-reference
• Docs = intended design (when code conflicts, docs win)
• Templates required before any Quoth updates

Full mode (~500 tokens) adds detailed workflows, trust level examples, and template mappings.

3. Pre-Edit/Write Reminders

Before Claude edits or creates code files, a lightweight reminder appears:

Quoth patterns available via `quoth_guidelines()` and `quoth_search_index`

~15 tokens. Just a nudge—Claude decides whether to use it.

4. Badge Enforcement

At the end of each response, if Claude used any Quoth tools, it displays a badge:

┌─────────────────────────────────────────────────┐
│ 🪶 Quoth                                        │
│   ✓ patterns/testing-pattern.md (vitest mocks) │
│   ✓ patterns/error-handling.md (try-catch)     │
└─────────────────────────────────────────────────┘

No Quoth usage? No badge. Clean response.

Token Comparison

92% reduction in hook overhead. 92% reduction in full guidelines size.

Why "Strongly Suggest" Instead of Force?

We experimented with mandatory searches before every edit. It works, but:

1. Adds friction for simple fixes where Quoth isn't relevant
2. Wastes tokens when Claude already knows the pattern
3. Slows down trivial tasks

The new model trusts Claude's judgment. The hints are clear. The tool is discoverable. Claude uses it when the task warrants it.

Installing the Plugin

Full Install (MCP + Hooks)

# Add the Quoth marketplace (one time)
/plugin marketplace add Montinou/quoth-mcp

# Install the plugin
/plugin install quoth@quoth-marketplace

This installs everything: MCP server, lightweight hooks, and the /quoth-genesis skill.

MCP Only (No Hooks)

If you prefer just the MCP tools without the hook reminders:

claude mcp add --transport http quoth https://quoth.ai-innovation.site/api/mcp

What About the Old Personas?

The /prompt quoth_architect, /prompt quoth_auditor, and /prompt quoth_documenter commands still work. They're useful when you want to enforce a strict behavioral mode for an entire conversation.

But for most workflows, quoth_guidelines('code') gives you the same rules with:

• Lower token cost
• More flexibility
• Better Claude autonomy

The Feather Badge 🪶

Why a feather? Quoth references Edgar Allan Poe's raven—"Quoth the Raven, Nevermore." The feather symbolizes the documentation patterns that keep your codebase consistent.

When you see 🪶 in a response, you know:

• Claude consulted your documentation
• Specific patterns were applied
• The code follows your team's standards

---

The Quoth Plugin is available now. Connect your MCP server and start your next Claude Code session to see it in action.