renoun-mcp 1.1.0

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

ReNoUn MCP Server

Structural observability for AI conversations. 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.

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

Tools

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

Quick Start

# Install dependencies
pip install numpy

# Optional: enables full MCP protocol (falls back to JSON-RPC stdio without it)
pip install mcp

# Place core.py in one of these locations:
#   1. Same directory as server.py
#   2. Parent directory
#   3. ~/.renoun/core.py
#   4. Set RENOUN_CORE_PATH=/path/to/core.py

# Run
python3 server.py

Integration

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

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

Claude Code

claude mcp add renoun python3 /absolute/path/to/renoun-mcp/server.py

Generic MCP Client

{
    "transport": "stdio",
    "command": "python3",
    "args": ["server.py"],
    "cwd": "/absolute/path/to/renoun-mcp/"
}

Environment Variable Configuration

# Point to core engine if not co-located
export RENOUN_CORE_PATH=/path/to/core.py

# Or use config file: ~/.renoun/config.json
# { "core_path": "/path/to/core.py" }

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.

What You Get Back

Health Score (DHS)

A 0.0–1.0 structural health metric. Below 0.45 = stuck or fragmenting. 0.55–0.75 = healthy movement. Above 0.75 = strong convergence.

8 Constellation Patterns

Each detected pattern includes an agent_action telling your agent what to do:

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

17 Channels

Five recurrence channels (stability), six novelty channels (disruption), six unity channels (coherence). Full breakdown available in references/CHANNELS.md.

Longitudinal Storage

Results persist to ~/.renoun/history/. Use renoun_pattern_query to:

• save: Store an analysis result with domain/tags
• list: See all stored sessions
• query: Filter by date, domain, constellation, DHS threshold
• trend: Compute health trajectory over time

Version

• Server: 1.1.0
• Engine: 4.1
• Schema: 1.1
• Protocol: MCP 2024-11-05 (or JSON-RPC stdio fallback)

Related

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

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