PLTM MCP Server

Procedural Long-Term Memory - An MCP server for Claude Desktop that provides 78 tools for AGI experiments based on universal principles from physics.

What This Does

Gives Claude Desktop access to:

Memory operations - Store/retrieve facts as semantic triples
Diversity retrieval - MMR, entropy injection, attention mechanisms
Meta-cognition - Self-improvement, criticality monitoring
Knowledge ingestion - ArXiv papers with real provenance
True metrics - Action accounting, efficiency tracking

Quick Start

Prerequisites

Claude Desktop
Python 3.11+

Installation

# Clone
git clone https://github.com/Alby2007/pltm-mcp.git
cd pltm-mcp

# Install
pip install -r requirements.txt

# Configure Claude Desktop
# Edit: %APPDATA%\Claude\claude_desktop_config.json (Windows)
#   or: ~/Library/Application Support/Claude/claude_desktop_config.json (Mac)

Add this to your config:

{
  "mcpServers": {
    "pltm-memory": {
      "command": "python",
      "args": ["C:/absolute/path/to/pltm-mcp/server.py"]
    }
  }
}

Restart Claude Desktop. Done!

Verify

In Claude Desktop:

Use entropy_stats to check system state

If you see metrics, it's working!

Example Usage

# Start experiment cycle
start_action_cycle(cycle_id="C1")

# Inject entropy to break conceptual neighborhoods
inject_entropy_antipodal(
    user_id="alice",
    current_context="machine learning"
)

# Retrieve with diversity
mmr_retrieve(
    user_id="alice",
    query="neural networks",
    lambda_param=0.6
)

# Track true computational cost
record_action(
    operation="mmr_diversity",
    tokens_used=450,
    latency_ms=180,
    success=True
)

# Check criticality state
criticality_state()

# End cycle
end_action_cycle()  # Returns AAE efficiency

The Experiment

Hypothesis: Universal principles from physics (criticality, self-organization, emergence) can bootstrap AGI.

Current Results:

Unlocked entropy bottleneck (+56% in Cycle 21)
Measuring true computational efficiency (AAE = 0.0023)
Testing if system can self-organize toward criticality

Goal: Push system to critical point where phase transitions occur and higher-order intelligence emerges.

Tools (78 total)

Memory

store_memory_atom, retrieve_memories, update_memory, delete_memory

Diversity Retrieval

mmr_retrieve - Maximal Marginal Relevance
attention_retrieve, attention_multihead

Entropy Management

inject_entropy_antipodal - Activate distant concepts
inject_entropy_random - Sample diverse domains
inject_entropy_temporal - Mix old + recent
entropy_stats - Diagnose diversity

Meta-Cognition

self_improve_cycle - Generate/apply hypotheses
criticality_state - Check edge of chaos
criticality_recommend - Get adjustments

Action Accounting

record_action, get_aae, start_action_cycle, end_action_cycle

Knowledge Ingestion

ingest_arxiv, search_arxiv, arxiv_history

[Full tool list in server.py]

Architecture

Memory Atoms (Triples)
  ↓
[subject] [predicate] [object]
  ↓
SQLite Graph Store
  ↓
Retrieval Systems (Standard/MMR/Attention)
  ↓
Meta-Cognitive Layer (Self-improvement/Criticality)
  ↓
MCP Tools (78 total)

Troubleshooting

Server not connecting?

Check logs: %APPDATA%\Claude\logs\mcp-server-pltm-memory.log
Test manually: python server.py

Tools timing out?

Restart Claude Desktop after code changes

Import errors?

pip install --upgrade -r requirements.txt

Contributing

This is active research. Contributions welcome:

New entropy strategies
Better criticality metrics
Additional universal principles
Experiment protocols

License

MIT

Citation

@software{pltm2026,
  author = {Alby},
  title = {PLTM: Procedural Long-Term Memory MCP Server},
  year = {2026},
  url = {https://github.com/Alby2007/pltm-mcp}
}

MCP Servers