GenPark Context Window Token Pruner

A local stdio MCP server that prunes LLM context windows with priority weighting and FIFO eviction to keep prompts within bounds.

What you can do

• Applies priority-weighted pruning to context content so lower-priority material is dropped first.
• Evicts older entries in first-in-first-out order to keep prompts within a token budget.
• Protects the LLM prompt boundary by trimming context before it is sent.
• No tool list, instructions or README are published, so the exact operations the assistant can call are not documented.

When to use it

Worth considering when long conversations or large retrieved context regularly overflow the model's context window and you want a local, deterministic trimming step rather than ad hoc summarization.

Requirements

Runs as a local process over stdio, distributed as an MCP bundle (mcpb) package; desktop-only, not executable in a web client. No authentication, environment variables or headers are declared.