MCP 服务器

renoun-mpc

Structural observability for AI conversations. Detects loops, stuck states, breakthroughs, and convergence across 17 channels without analyzing content.

README

<h1 align="center">ReNoUn</h1> Structural observability for AI conversations <a href="https://pypi.org/project/renoun-mcp/"><img src="https://img.shields.io/pypi/v/renoun-mcp?color=7C9A6E&label=PyPI" alt="PyPI"></a> <a href="https://pypi.org/project/renoun-mcp/"><img src="https://img.shields.io/pypi/pyversions/renoun-mcp?color=5B7B9E" alt="Python"></a> <a href="https://github.com/98lukehall/renoun-mcp/blob/main/LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue" alt="License"></a> <a href="https://web-production-817e2.up.railway.app/docs"><img src="https://img.shields.io/badge/API-docs-orange" alt="API Docs"></a>

Your agent doesn't know when it's going in circles. ReNoUn does.

Detects when conversations are stuck in loops, producing cosmetic variation instead of real change, or failing to converge. Measures structural health across 17 channels without analyzing content — works on any turn-based interaction.

Why?

LLMs get stuck. They produce responses that sound different but are structurally identical — what we call surface variation. A human might notice after 5 turns. An agent never will.

ReNoUn catches this in ~200ms by measuring structure, not content. It works on any language, any topic, any model.

Install

pip install renoun-mcp

Quick Start

As an MCP Server (Claude Desktop)

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
    "mcpServers": {
        "renoun": {
            "command": "python3",
            "args": ["-m", "server"],
            "env": {
                "RENOUN_API_KEY": "rn_live_your_key_here"
            }
        }
    }
}

As a REST API

curl -X POST https://web-production-817e2.up.railway.app/v1/analyze \
  -H "Authorization: Bearer rn_live_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{"utterances": [
    {"speaker": "user", "text": "I feel stuck"},
    {"speaker": "assistant", "text": "Tell me more about that"},
    {"speaker": "user", "text": "I keep going in circles"},
    {"speaker": "assistant", "text": "What patterns do you notice?"},
    {"speaker": "user", "text": "The same thoughts repeat"}
  ]}'

As a Claude Code MCP

claude mcp add renoun python3 -m server

Demo Output

{
  "dialectical_health": 0.491,
  "loop_strength": 0.36,
  "channels": {
    "recurrence": { "Re1_lexical": 0.0, "Re2_syntactic": 0.3, "Re3_rhythmic": 0.5, "Re4_turn_taking": 1.0, "Re5_self_interruption": 0.0, "aggregate": 0.36 },
    "novelty":    { "No1_lexical": 1.0, "No2_syntactic": 1.0, "No3_rhythmic": 0.5, "No4_turn_taking": 0.5, "No5_self_interruption": 0.0, "No6_vocabulary_rarity": 0.833, "aggregate": 0.639 },
    "unity":      { "Un1_lexical": 0.5, "Un2_syntactic": 0.135, "Un3_rhythmic": 0.898, "Un4_interactional": 0.7, "Un5_anaphoric": 0.705, "Un6_structural_symmetry": 0.5, "aggregate": 0.573 }
  },
  "constellations": [],
  "novelty_items": [
    { "index": 4, "text": "The same thoughts repeat", "score": 0.457, "reason": "shifts conversational direction" }
  ],
  "summary": "Moderate dialectical health (DHS: 0.491). Diverse exploration (loop strength: 0.36). Key moment at turn 4.",
  "recommendations": ["■ Key novelty at turn 4. Consider returning to this moment."]
}

Tools

Tool	Purpose	Speed	Tier
`renoun_analyze`	Full 17-channel structural analysis with breakthrough detection	~200ms	Pro
`renoun_health_check`	Quick triage — one score, one pattern, one action	~50ms	Free
`renoun_compare`	Structural A/B test between two conversations	~400ms	Pro
`renoun_pattern_query`	Save, query, and trend longitudinal session history	~10ms	Pro

How It Works

ReNoUn measures 17 structural channels across three dimensions:

Recurrence (5 channels) — Is structure repeating? Lexical, syntactic, rhythmic, turn-taking, and self-interruption patterns.

Novelty (6 channels) — Is anything genuinely new emerging? Lexical novelty, syntactic novelty, rhythmic shifts, turn-taking changes, self-interruption breaks, and vocabulary rarity.

Unity (6 channels) — Is the conversation holding together? Lexical coherence, syntactic coherence, rhythmic coherence, interactional alignment, anaphoric reference, and structural symmetry.

From these 17 signals, ReNoUn computes a Dialectical Health Score (DHS: 0.0–1.0) and detects 8 constellation patterns, each with a recommended agent action:

Pattern	What It Means	Agent Action
CLOSED_LOOP	Stuck recycling the same structure	`explore_new_angle`
HIGH_SYMMETRY	Rigid, overly balanced exchange	`introduce_variation`
CONVERGENCE	Moving toward resolution	`maintain_trajectory`
PATTERN_BREAK	Something just shifted	`support_integration`
SURFACE_VARIATION	Sounds different but structurally identical	`go_deeper`
SCATTERING	Falling apart, losing coherence	`provide_structure`
REPEATED_DISRUPTION	Keeps breaking without stabilizing	`slow_down`
DIP_AND_RECOVERY	Disrupted then recovered	`acknowledge_shift`

Pricing

	Free	Pro ($4.99/mo)
`renoun_health_check`	✓	✓
`renoun_analyze`	—	✓
`renoun_compare`	—	✓
`renoun_pattern_query`	—	✓
Daily requests	20	1,000
Max turns per analysis	200	500

Get your API key: Subscribe via Stripe or visit harrisoncollab.com.

REST API

Base URL: https://web-production-817e2.up.railway.app

Endpoint	Method	Auth	Description
`/v1/analyze`	POST	Bearer	Full 17-channel analysis
`/v1/health-check`	POST	Bearer	Fast structural triage
`/v1/compare`	POST	Bearer	A/B test two conversations
`/v1/patterns/{action}`	POST	Bearer	Longitudinal pattern history
`/v1/status`	GET	None	Liveness + version info
`/v1/billing/checkout`	POST	None	Create Stripe checkout session
`/docs`	GET	None	Interactive API explorer

All authenticated endpoints require: Authorization: Bearer rn_live_...

Input Format

All analysis tools accept conversation turns as speaker/text pairs:

{
    "utterances": [
        {"speaker": "user", "text": "I keep going back and forth on this decision."},
        {"speaker": "assistant", "text": "What makes it feel difficult to commit?"},
        {"speaker": "user", "text": "I think I'm afraid of making the wrong choice."}
    ]
}

Minimum 3 turns required. 10+ recommended for reliable results. 20+ for stable constellation detection.

Integration

Claude Desktop

{
    "mcpServers": {
        "renoun": {
            "command": "python3",
            "args": ["-m", "server"],
            "env": { "RENOUN_API_KEY": "rn_live_your_key_here" }
        }
    }
}

Claude Code

RENOUN_API_KEY=rn_live_your_key_here claude mcp add renoun python3 -m server

Generic MCP Client

{
    "transport": "stdio",
    "command": "python3",
    "args": ["-m", "server"],
    "env": { "RENOUN_API_KEY": "rn_live_your_key_here" }
}

Environment Variable

export RENOUN_API_KEY=rn_live_your_key_here

Longitudinal Storage

Results persist to ~/.renoun/history/. Use renoun_pattern_query to save, list, query, and trend session history over time. Filter by date, domain, constellation pattern, or DHS threshold.

Version

Server: 1.2.0
Engine: 4.1
Schema: 1.1
Protocol: MCP 2024-11-05

The ReNoUn Cowork Plugin provides skill files, slash commands, and reference documentation for agents using the Cowork plugin system. The MCP server and plugin share the same engine and can be used independently or together.

Patent Notice

The core computation engine is proprietary and patent-pending (#63/923,592). This MCP server wraps it as a black box. Agents call engine.score() and receive structured results — they never access internal algorithms.

License

MCP server and API wrapper: MIT. Core engine: Proprietary.

<a href="https://harrisoncollab.com">Harrison Collab</a> · <a href="https://web-production-817e2.up.railway.app/docs">API Docs</a> · <a href="https://pypi.org/project/renoun-mcp/">PyPI</a>