PromptThrift MCP
JSON →Smart token compression for LLM apps. Save 70-90% on API costs with Gemma 4 local compression, multi-model cost tracking, and intelligent model routing.
Install
pip install (recommended)** Tools · 4
- promptthrift_compress_history Compress old turns into a smart summary to reduce input tokens by 50-90%
- promptthrift_count_tokens Track token usage and costs across 14 models
- promptthrift_suggest_model Recommend the cheapest model for a given task to save 60-80% on simple tasks
- promptthrift_pin_facts Pin critical facts that survive compression to never lose key context
Environment variables
PROMPTTHRIFT_OLLAMA_URL
Links
★ 1 GitHub stars