Context Window Eviction Planner

Context Window Eviction Planner MCP Connector for Claude

A+

A deterministic engine to manage LLM context windows by calculating optimal message eviction.

3 tools Official Updated Oct 1, 2026 Official Vinkius Partner

This MCP server provides precise tools for managing LLM context limits. It calculates an exact eviction plan to ensure conversations stay within technical bounds while minimizing information loss. Use calculate_eviction_plan to determine which messages to remove based on relevance and age, or analyze_window_efficiency to monitor context pressure. It is designed to prevent context overflow and suggest summarization when significant data loss is imminent.

context-windowtoken-managementllm-optimizationeviction-logicefficiency

3 tools expose this connector's capabilities to your AI agent.

get_message_metadata

Provides a summary of the importance and "weight" of the current conversation history

analyze_window_efficiency

Evaluates how effectively the current context is being used and flags high-risk scenarios

calculate_eviction_plan

Determines exactly which messages to remove to fit the conversation within the available context window

See how to talk to your AI agent using Context Window Eviction Planner.

Calculate an eviction plan for a 4000 token window with 3500 tokens currently used.

The eviction plan has been generated. 2 messages were removed to bring the total usage down to 3200 tokens, resulting in an information loss of 0.15.

Check the efficiency of my current context window.

The current context utilization is 85%, and the status is healthy.

What is the total token count and average relevance of my messages?

The conversation contains 45 messages with a total of 2800 tokens and an average relevance score of 0.72.

Priority is calculated using the inverse of a message's relevance score and its age. Messages with low relevance and high age are prioritized for removal via `calculate_eviction_plan`.

Related Connectors