PLTM

You can extend Claude Desktop with the PLTM MCP Server to access memory, diversity-based retrieval, meta-cognition, knowledge ingestion, and precise action accounting through a dedicated MCP. This server enables experimentation with universal physics-inspired principles to bootstrap advanced reasoning and self-improvement workflows within Claude Desktop.

How to use

To use the PLTM MCP Server, first ensure Claude Desktop is installed and your environment has Python 3.11 or newer. You will run the MCP as a local process and connect Claude Desktop to it via the provided configuration. Once connected, you can perform memory operations, leverage diverse retrieval strategies, monitor meta-cognition metrics, ingest knowledge from sources like ArXiv, and track true computational costs during experiments.

How to install

Prerequisites: You need Claude Desktop and Python 3.11 or newer.

Step 1: Clone the repository and enter the project directory.

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

Step 2: Install the Python dependencies.

pip install -r requirements.txt

Step 3: Configure Claude Desktop with the MCP server recipient. On Windows, edit the config at %APPDATA%\Claude\claude_desktop_config.json. On macOS, edit ~/Library/Application Support/Claude/claude_desktop_config.json.

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

Start and verify connectivity

Restart Claude Desktop after configuring the MCP entry. To verify that the MCP is running and reachable, use the system utilities for your environment to confirm the MCP process is active and then issue a diagnostic query from Claude Desktop to check the entropy or system state, such as requesting entropy statistics.

Verification steps after setup

In Claude Desktop, run a quick check to confirm metrics are available. If entropy statistics and related metrics appear, the MCP connection is functioning correctly.

Available tools

store_memory_atom

Store a memory atom as a semantic triple (subject, predicate, object) and index it in the memory store.

retrieve_memories

Retrieve stored memory triples matching given criteria to support retrieval-based reasoning.

update_memory

Update an existing memory atom to refine or correct stored information.

delete_memory

Delete a memory atom from the store.

mmr_retrieve

Perform Maximal Marginal Relevance retrieval to fetch diverse results for a query.

attention_retrieve

Retrieve items using attention-based ranking to emphasize contextually relevant results.

attention_multihead

Use multi-head attention strategies to improve retrieval diversity and relevance.

inject_entropy_antipodal

Activate distant concepts to expand exploration and cross-domain connections.

inject_entropy_random

Sample diverse domains randomly to broaden the search space.

inject_entropy_temporal

Mix old and recent information to maintain a dynamic knowledge horizon.

entropy_stats

Diagnose diversity and entropy-related properties of the knowledge base.

self_improve_cycle

Generate and apply hypotheses for self-improvement cycles.

criticality_state

Check the system’s position relative to the edge of chaos to gauge readiness for higher-order behavior.

criticality_recommend

Provide adjustments and actions to steer the system toward criticality.

record_action

Log an operation with metadata such as tokens used and latency, for auditing and cost tracking.

get_aae

Retrieve the accumulated action efficiency metrics (AAE) for evaluation.

start_action_cycle

Begin a new action cycle and assign identifiers for tracking.

end_action_cycle

Conclude an action cycle and return the resulting efficiency metrics.

ingest_arxiv

Ingest ArXiv papers with provenance for knowledge expansion.

search_arxiv

Query ArXiv to locate relevant literature for ingestion.

arxiv_history

Fetch provenance/history of ingested ArXiv items.