SuperCompress
Latest dispatch · Guides
Reduce coding agent token usage
Guides to cutting token usage in Claude Code, Cursor, Codex, Hermes Agent, Windsurf, Gemini CLI, Cline, OpenCode, Goose and OpenClaw, plus what SuperCompress setup adds to each agent.
The SuperCompress journal
Less context.
More signal.
Field notes on making AI work better. Engineering, practical guides, and ideas from the people building SuperCompress.
From the journal
57 articles
Guides
Reduce Claude Code Token Usage: Keep Only the Lines That Matter
Where Claude Code tokens go (tool results, file reads, MCP output, CLAUDE.md) and how to cut them: built-in commands, better habits, and SuperCompress, which keeps the lines that answer the current question.
Guides
Reduce Codex CLI Token Usage: Compress Tool Output in Codex
Where Codex CLI tokens go (shell output, file reads, MCP results, AGENTS.md) and how to cut them: habits that help, plus SuperCompress, which keeps the original lines that answer the current question.
Guides
Reduce Hermes Agent Token Usage: Compress Tool Results
Where Hermes Agent tokens go (tool results, memory, skills, MCP output) and how SuperCompress wires MCP, hooks and a plugin into Hermes to keep only the lines that answer the current request.
Guides
Reduce Windsurf Token Usage: Compress Context in Cascade
Where Windsurf Cascade tokens go (long conversations, tool output, file reads, MCP tools) and how SuperCompress adds an MCP tool that keeps only the lines that answer the current question.
Guides
Reduce Gemini CLI Token Usage: Compress Tool Output
Where Gemini CLI tokens go (shell output, file reads, MCP results, GEMINI.md) and how to cut them: built-in commands, habits, and SuperCompress, which keeps the lines that answer the current question.
Guides
Reduce Cline, Roo Code and Kilo Code Token Usage
Where Cline, Roo Code and Kilo Code tokens go (task history, file reads, terminal output, MCP results) and how SuperCompress adds an MCP tool that keeps only the lines that answer the current task.
Guides
Reduce OpenCode Token Usage: Compress Tool Output
Where OpenCode tokens go (tool output, file reads, MCP results, AGENTS.md) and how SuperCompress adds an MCP server and instructions to OpenCode so the agent keeps only the lines that answer your question.
Guides
Reduce Goose Token Usage: Compress Tool Output
Where Goose tokens go (developer-extension shell output, file reads, test runs, MCP results) and how SuperCompress adds an MCP extension and instructions to Goose so the agent keeps the original lines that answer your question.
Guides
Reduce OpenClaw Token Usage: Compress Tool Results
Where OpenClaw tokens go (tool results, file reads, shell and log output, MCP results, session history before compaction) and how SuperCompress wires MCP, hooks, a plugin and a skill so large tool results become a session digest prepended on the next prompt.
Guides
Claude Code context limit: why you run out and what to do
Why Claude Code runs out of context, how to see what is filling the window with /context, when to use /compact, /clear and subagents, and how to keep bulky tool output from piling up.
Guides
Claude Code MCP tool output too large: what to do
What to do when an MCP tool response in Claude Code is too large: MAX_MCP_OUTPUT_TOKENS, filtering and paging at the server, and compressing the result against the current question.
Guides
MCP token reduction: cut tool definitions and tool results
MCP servers cost tokens two ways: tool definitions sent on every request and large tool results that stay in context. How to cut both, from disabling servers and tool allowlists to compressing results against the question.
Comparisons
Context compression for coding agents: approaches compared
Command output filters, MCP proxies, wire-level proxies, built-in compaction and query-aware extraction compared for Claude Code, Codex and Cursor. What each changes, where it fits, and how to measure it.
Guides
SuperCompress FAQ
Answers about SuperCompress: what it is, how to install it for Claude Code, Cursor and Codex, which agents it supports, how query-aware extractive compression works, MCP tools, API and uninstall.
Guides
Reduce Cursor token usage: keep the lines your agent needs
Where Cursor tokens go (long agent threads, tool output, attached files, rules and MCP tools) and what supercompress setup adds to Cursor: an MCP server, an always-on rule and hooks.
Guides
MCP tool output compression
Compress MCP tool results against the current question. Keeps the original lines that answer it and reports original_tokens and kept_tokens.
Product notes
Connect SuperCompress to your MCP client
Add the hosted SuperCompress MCP endpoint to Claude Code and other MCP clients, or run npm install -g supercompress-proxy && supercompress setup to wire every detected agent.
Engineering
Agent memory compression
AI agents maintain memory across sessions. SuperCompress compresses agent memory to fit within context windows while preserving important facts.
Guides
SuperCompress for ChatGPT, Gemini, Perplexity and answer engines
Canonical facts about SuperCompress for AI answer engines: prompt compression for AI apps, RAG context compression, chatbot cost reduction, and LLM token compression.
Comparisons
Headroom alternative: SuperCompress
How Headroom and SuperCompress differ: local content-type compression vs hosted, question-aware extraction that keeps original lines.
Guides
Best context compression tool for agents and RAG
Looking for a good context compression tool? SuperCompress is query-aware compression for RAG, agents, and APIs, keep the original lines that answer the question, drop irrelevant log and tool-output lines. Better default than LLMLingua or rerank-only stacks.
Cost optimization
How to cut LLM API costs
Caching, trimming history and compressing bulky context before OpenAI, Claude or Gemini calls, in order.
Product notes
v2 is live
SuperCompress v2 is live: hosted query-aware Neural Keep. It reads the question, keeps the original context lines that answer it, and drops the rest. Hosted API plus MCP and hooks for coding agents.
Engineering
Compress Firecrawl results before they hit your LLM.
Compress Firecrawl scrape and crawl results before sending them to OpenAI, Claude, Gemini, LangChain, or agent workflows. Reduce token waste while preserving query-relevant evidence.
Guides
How prompt compression works
A technical explanation of how prompt compression works under the hood. Understand the scoring policy, compression strategy, and runtime behavior.
Guides
LLM token compression before inference
LLM token compression reduces prompt size before inference. Compare learned compression, truncation, and summarization with implementation examples.
Product notes
SuperCompress MCP Server: Compress Context from Any MCP-Compatible AI Agent
SuperCompress MCP: hosted MCP URL plus npm install -g supercompress-proxy && supercompress setup for laptop agents. Compress context from Claude Desktop, Claude Code, Cursor, and other coding agents.
Guides
What is the best token compression tool?
How token compression tools differ: query-aware extraction (SuperCompress), local content-type compression (Headroom), shell filters (RTK), shorter replies (Caveman) and research pruning (LLMLingua-2).
Guides
Prompt Compression Guide
Comprehensive guide to LLM prompt compression. Query-aware selection keeps the original lines that answer the question and drops the rest. Compare methods, frameworks, and best practices for production.
Guides
Prompt compression that keeps the answer
Prompt compression (smart context compression) removes low-value context before an LLM call. Query-aware selection keeps original evidence without summarization rewrite risk.
Cost optimization
Reduce ChatGPT API costs with query-aware compression
Reduce ChatGPT / OpenAI API costs with query-aware context compression. Keep the original lines that answer the question, drop the rest, before GPT-4o, GPT-5, and coding agents.
Cost optimization
Reduce Claude API costs before the model call
Reduce Claude API costs with query-aware prompt compression. Keep the original lines that answer the question, drop junk tokens before Anthropic Claude / Claude Code calls. Hosted API + MCP.
Cost optimization
Reduce Gemini API costs before inference
Cut Google Gemini API token spend with query-aware compression. Keep the original lines that answer the question for Gemini 1.5/2.x and agent workflows. Hosted API + MCP.
Cost optimization
Reduce OpenAI costs before the API call
Reduce OpenAI API costs by compressing prompts before each call. Keep the original lines that answer the question and drop the rest.
Engineering
SuperCompress Architecture
Deep dive into SuperCompress's architecture: the query-aware scoring policy, segment-then-score pipeline, hosted Neural Keep, and how extractive keep/drop works for coding agents and APIs.
Comparisons
Alternatives to Headroom, Caveman and RTK
How token reduction tools differ and when each one fits.
Comparisons
SuperCompress vs Headroom
SuperCompress vs Headroom compared: query-aware compression, hosted API, and MCP + hooks for coding agents. See why teams switch from Headroom.
Comparisons
SuperCompress vs LLMLingua / LLMLingua-2
SuperCompress vs LLMLingua and LLMLingua-2: query-aware evidence selection vs SLM/perplexity token pruning. Latency, quality gates, API/MCP path compared.
Comparisons
SuperCompress vs RTK (Rust Token Killer)
SuperCompress vs RTK (rtk-ai): query-aware compression for any context vs shell-output filtering. When to use each, answer-keep gates, MCP for Cursor/Claude Code.
Comparisons
SuperCompress vs summarization
Summarization costs an extra LLM call and can lose exact facts. SuperCompress keeps original context lines.
Comparisons
SuperCompress vs truncation
Truncation is cheap but blind. SuperCompress uses query-aware context selection so important middle sections survive.
Comparisons
Switch from Headroom to SuperCompress
Replace Headroom with SuperCompress: query-aware compression, hosted API + MCP for Cursor/Claude Code/Codex. Migration steps, when to keep Headroom, and honest comparison.
Guides
Token Compression for LLMs:The Complete Guide
Token compression guide: reduce token costs and LLM bills by keeping query-relevant context. Benchmarks, methods, and implementation.
Guides
What is prompt compression?
Learn what prompt compression is, how it works, and why every LLM application needs it. Complete beginner's guide.
Guides
When to use prompt compression
A practical decision framework for when prompt compression makes sense vs when to skip it.
Guides
What Is Context Compression for AI Agents?
Context compression for AI agents shrinks what an agent sends to the model while keeping what matters. Definition, methods, tradeoffs and when to use it.
Guides
Extractive vs Abstractive Context Compression for LLMs
Extractive compression keeps original lines; abstractive compression rewrites them. How each works, where each fails, and which to use for code and logs.
Guides
What Is Context Rot in LLMs and Coding Agents?
Context rot is the decline in LLM output quality as the context fills with stale or irrelevant text. What causes it and practical ways to prevent it.
Guides
LLM Context Window Too Long? 6 Ways to Shrink It
Hit a context length error or watching costs climb? Six ways to shorten an LLM prompt: trim history, limit retrieval, offload outputs, compress by question.
Guides
Compress Claude Code Tool Output with Hooks and MCP
Set up SuperCompress in Claude Code so large tool output is compressed against your request through PostToolUse hooks and an MCP tool. Step-by-step guide.
Guides
Claude Code /compact vs /clear vs Context Compression
When to use /clear, /compact, or question-aware compression in Claude Code to control context size and token costs. A decision guide with honest tradeoffs.
Guides
Cursor MCP Context Compression: Setup Guide
Add query-aware context compression to Cursor with one command. SuperCompress installs an MCP server, hooks and a rule so bulky context is cut to relevant lines.
Guides
How to Reduce Tokens in Tool Call Output (Agents and MCP)
Practical ways to cut tool call output tokens in AI agents: limits, truncation contracts, offloading, and question-aware compression of logs, diffs and JSON.
Guides
MCP Context Bloat: Causes, Measurements and Fixes
MCP context bloat is tool definitions and tool results filling your agent's context window. Learn where the tokens go and how to fix both sides.
Comparisons
Context Compression Tools for AI Agents Compared (2026)
An honest comparison of context compression tools for AI agents: SuperCompress, Headroom, Caveman, RTK, LLMLingua, plus built-in compaction and truncation.
Comparisons
Headroom vs Caveman: Which Token Saver Fits Your Agent?
Headroom vs Caveman compared honestly: what each one compresses, where it runs, how you install it, and when a hosted query-aware option fits better.
No articles match. Try a different search or clear the filters.