Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@genpark-contextual-memory-pruning-importance-attributor-skill Score and prune my chat memories to fit 500 tokens, keeping pinned items."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Autonomous Long-Horizon Agentic Memory Pruning & Importance Attributor. Evaluates multi-turn episodic memory buffers, calculates multi-factor salience scores (recency decay, user priority pinning, emotional valence, and entity graph connectivity), and deterministically prunes low-utility noise to respect model context limits.
100% Zero External Dependencies: Runs entirely on the Python 3.9+ standard library.
Model Context Protocol (MCP) Standard: Native support for JSON-RPC 2.0 initialize, tools/list, and tools/call.
Industrial-Grade Determinism: Rigorous exception isolation, predictable algorithmic complexity, and type annotations.
Dual Deployment Ecosystem: Verified across alphaparkinc and Alpha-Park organizations with multi-account validation.
Related MCP server: agent-memory
"""Example usage for ContextualMemoryPruningImportanceAttributor."""
import sys
import json
import time
from client import ContextualMemoryPruningImportanceAttributor
sys.stdout.reconfigure(encoding='utf-8')
def main():
print("=== Long-Horizon Contextual Memory Pruning Demo ===")
attributor = ContextualMemoryPruningImportanceAttributor()
now = time.time()
memory_pool = [
{
"id": "mem_01",
"content": "User account ID: ACC-99412, Stripe Customer ID: CUST-STRIPE-7788",
"timestamp": now - 7200,
"access_count": 28,
"is_pinned": True
},
{
"id": "mem_02",
"content": "User mentioned liking black coffee with oat milk on Monday morning.",
"timestamp": now - 86400 * 5,
"access_count": 2,
"is_pinned": False
},
{
"id": "mem_03",
"content": "Autonomous deployment target: Tencent Cloud CVM cluster cvm-sh-prod-02.",
"timestamp": now - 1800,
"access_count": 14,
"is_pinned": False
},
{
"id": "mem_04",
"content": "Random greeting: 'Hey agent how are you today doing fine'.",
"timestamp": now - 86400 * 10,
"access_count": 1,
"is_pinned": False
}
]
print("\n--- 1. Evaluating Multi-Factor Importance Attribution ---")
for m in memory_pool:
score = attributor.attribute_memory_importance(m)
print(f"[{m['id']}] Score: {score:.3f} | Pinned: {m.get('is_pinned', False)} | Text: '{m['content'][:45]}...'")
print("\n--- 2. Pruning Memories to Strict Token Budget (40 Tokens) ---")
pruned_res = attributor.prune_contextual_memories(memory_pool, target_token_budget=40)
print(f"Initial Tokens: {pruned_res['initial_tokens']} -> Retained Tokens: {pruned_res['retained_tokens']}")
print(f"Savings: {pruned_res['token_savings_pct']}% | Retained Nodes: {pruned_res['retained_memory_count']}/{pruned_res['initial_memory_count']}")
print("Retained Memory IDs:", [m["id"] for m in pruned_res["retained_memories"]])
print("Pruned Memory IDs:", pruned_res["pruned_memory_ids"])
if __name__ == "__main__":
main()
Start standard JSON-RPC 2.0 server over stdio:
python mcp_server.pyExecute embedded test harness:
python mcp_server.py --testInspect skill.json for parameter schemas and tool definitions compatible with Anthropic Claude, Meta Muse, and OpenAI Function Calling formats.
Licensed under the MIT License. Copyright © 2026 GenPark AI.
This server cannot be deployed
Agent memory that refuses to guess: evidence-gated recall, exact-source reads, verifiable deletion.
Persistent memory for AI agents to retain, retrieve, and recall conversation context through MCP.
Hosted agent memory with provenance, contradiction surfacing, and snapshot rollback.
Memory that reasons: continual learning for stateful agents. Better context, fewer tokens.
