Agent♥︎Age
Catalog

io.github.JesusGarcia9009/lethe

Official

by JesusGarcia9009 · Python

Context GC for LLM agents: offload large tool outputs and recall them to save tokens.

Model Context Protocol (MCP) Server: io.github.JesusGarcia9009/lethe

The lethe MCP server provides “Context GC for LLM agents” that offloads large tool outputs out of an agent’s context and recalls them on demand. The goal is to reduce token usage for long tasks by managing ephemeral token and history data.

🛠️ Key Features

  • Context garbage collection for LLM agents
  • Offload large tool outputs from the active context
  • Recall stored outputs on demand
  • “Live Ephemeral Token & History Engine” per README excerpt

🚀 Use Cases

  • Long-running LLM agent workflows that produce large tool outputs
  • Token optimization during multi-step tool use
  • Context management for MCP-based tooling

⚡ Developer Benefits

  • Save tokens by removing large outputs from context until needed
  • Support model-context-protocol (MCP) integrations
  • Python-focused implementation (project topics)

⚠️ Limitations

  • Documentation excerpt is partial; detailed interface, configuration, and operational constraints are not provided in the source data.

Topics

ai-agentsanthropicclaude-codecodexcontext-managementcontext-windowgarbage-collectorllmmcpclaudellm-toolsmodel-context-protocolpythontoken-optimization