SuperCompress

io.github.Supercompressv0.5.38更新於 Oct 3, 2026

Compress coding-agent context ~64%. Hosted Neural Keep MCP for Cursor/Claude/Codex.

已驗證Streamable HTTP可網頁執行Developer ToolsAI & MLProductivity & Workflow

概覽

AI 產生的概覽

針對目前問題壓縮龐大的編碼代理上下文，以減少 LLM 輸入 token，並透過 MCP 提供壓縮、帳號連結與用量工具。

功能
SuperCompress 會針對目前問題對檔案、日誌、工具輸出與貼上內容等龐大上下文進行評分，丟棄填充內容，同時保留對證據關鍵的原始行。它提供 compress_context、connect_account 與 usage_summary 等 MCP 工具，並可向多種編碼工具安裝 hooks 與 MCP 設定。選用的本機代理可為相容 OpenAI 與 Anthropic 的用戶端重寫 base URL。
適用情境
當編碼代理經常傳送大型檔案、日誌或工具輸出，而你希望在不遺失關鍵答案內容的前提下減少輸入 token 時，值得加入。它適用於支援 MCP 或 hooks 的 Cursor、Claude Code、Codex 等工具。
執行需求
需要 Node.js 18+ 以執行 npm 套件 supercompress-proxy，可透過 npx 或全域安裝。需透過 setup 指令或 OAuth/裝置連結綁定 SuperCompress 帳號；選用 API 金鑰透過 SUPERCOMPRESS_API_KEY 傳入。需要連線至 SuperCompress API 的網路，選用本機代理會監聽 localhost:8080。
安裝前請注意
上下文文字會傳送至 SuperCompress API 進行託管壓縮，因此不要透過它傳送機密或受監管資料。服務按量計費：每月有免費 token 額度，超出後計費，可用 usage_summary 查看配額。setup 指令會為偵測到的代理寫入 MCP 與 hook 設定，uninstall 會刪除使用者家目錄下的檔案。請將 SUPERCOMPRESS_API_KEY 視為機密。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 SuperCompress，將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "supercompress": {
      "type": "http",
      "url": "https://api.supercompress.dev/api/mcp"
    }
  }
}

README

SuperCompress

SuperCompress v2 — cut ~64% of LLM input tokens for coding agents, without losing the answer.

~400M Neural Keep on the SuperCompress API scores bulky context (files, logs, tool dumps, pastes) against the current question and drops the rest. Your ask stays intact. This package wires Cursor / Claude Code / Codex / 60+ harnesses into that API via MCP + hooks.

Website · Benchmarks · Playground · Docs

Install

bash
npm install -g supercompress-proxy

Requires Node.js 18+.

Quick start (recommended)

One command links your account and auto-adds MCP + hooks across 60+ harnesses (25+ get a native auto MCP plugin — Cursor, Claude Code, Codex, Goose, Zed, OpenCode, fx, Hermes, OpenClaw, Grok, Gemini, Windsurf, Continue, and more):

bash
supercompress setupsupercompress doctor

Then restart your agent so integrations reload. That’s it.

Re-detect later (new agent installed, etc.):

bash
supercompress pluginsupercompress agentssupercompress doctor

Any other agent (custom harness, closed-source, DIY):

bash
supercompress agents connect

That prints a stdio MCP snippet and writes an Agent Plugins 1.0 pack you can drop into any compatible client.

Benchmarks

Same keep-budget (35% of tokens kept). Who still has the answer?

MethodAnswer-critical kept
FIFO / truncation24.8%
Summarization60.5%
H2O97.9%
SuperCompress100%
MetricResult
Oracle recall (fixed budget)100%
Mean token cut (real suite)~67%
Important lines kept (compiler)100%

Full methodology and charts: supercompress.dev/benchmarks

How it works

Your agent ──→ SuperCompress (hooks / MCP) ──→ smaller context ──→ model                      ↑                 query stays whole; only context is compressed
  1. You ask a question (never rewritten).
  2. Large context is scored against that question.
  3. Evidence-critical lines stay in original wording; filler drops.
  4. You pay for fewer input tokens.

Commands

CommandWhat it does
supercompress / tuiInteractive paper-branded UI (default in a TTY; Bun)
supercompress setupRecommended — link account, detect agents, install MCP + hooks
supercompress pluginRefresh agent integrations anytime
supercompress doctorPer-harness health matrix (account / MCP / hooks)
supercompress agents60+ catalog — auto-plugin vs recipe/connect
supercompress start / stop / statusOptional local proxy (setup --proxy)
supercompress usagePlan, quota, savings (--json ok)
supercompress uninstallRemove configs under ~/.supercompress

Optional localhost API proxy (base-URL rewrite) if you explicitly need it:

bash
supercompress setup --proxysupercompress start

Then point OpenAI/Anthropic-compatible clients at http://localhost:8080/v1.

MCP

setup / plugin registers the MCP server on every detected host. You can also run it directly:

bash
supercompress-mcp

Manual registration:

json
{  "mcpServers": {    "supercompress": {      "command": "supercompress-mcp"    }  }}
ToolPurpose
compress_contextCompress bulky context for a query
connect_accountLink this install to your dashboard
usage_summarySavings for the connected account

Account & pricing

Launch promo: 5M tokens/month free, then $0.10 / 1M from the dashboard.

Privacy

Hooks / MCP run on your machine. Provider API keys stay with your agent. Context text is sent to the SuperCompress API so the hosted compiler can compress it.

More

License

MIT — see LICENSE.

來源：packages/proxy/README.md，提交 08a34db

工具

0
工具後設資料尚未被收錄。

版本歷史

1
  1. v0.5.38最新Oct 3, 2026