gwen-digestor — Conversation compression MCP server

[NcrMancer]

gwen-digestor is an open-source MCP server that compresses conversation exchanges before they enter the LLM context window, reducing token consumption by ~38%.

Features

• 4 MCP tools: digest_input, compress_response, cache_reference, session_stats
• Mode-aware compression: auto-detects check-in, task, casual, and narrative messages
• Content-type detection: smart JSON crushing, code comment stripping, prose pass-through
• Deterministic: no embeddings, no GPU, no external API calls
• Gzip-compressed cache: SQLite-backed reference store with TTL expiry
• Stats tracking: persistent token savings measurement

Results

• 38.3% overall token reduction
• ~72% reduction on output responses

Links

• GitHub: https://github.com/NcrMancer/gwen-digestor
• Report with charts: https://github.com/NcrMancer/gwen-digestor/blob/main/docs/gwen-digestor-token-reduction-report.pdf