🔥 Context Mode：14.6k Stars 的上下文优化 MCP 服务器，98% Token 节省，15 个平台一把梭
Context Mode 是个 MCP 服务器，核心干一件事：管好 AI Coding Agent 的上下文窗口。一个 Playwright 截图 56 KB，20 个 GitHub Issue 59 KB，500 条访问日志 45 KB——30 分钟后 40% 的上下文就没了。Context Mode 的沙箱工具把 315 KB 原始输出压到 5.4 KB，省 98%。
不是滤掉输出，是压根不让原始数据进上下文。数据在沙箱子进程里处理完，只把摘要喂给 AI。
四个维度解决上下文问题
2. 会话连续性——每一次编辑、git 操作、任务、错误、用户决策都记到 SQLite。上下文压缩时不是把数据倒回来，而是建 FTS5 索引，BM25 检索，模型接着上次的问继续
3. Think in Code——别让 LLM 读 50 个文件去数函数行数，写个脚本算了就 console.log 结果。一个脚本替换十次工具调用，省 100 倍上下文
4. 不碰输出风格——只管数据去哪，不管模型怎么写回答
装一个试试（Claude Code）
# 插件市场安装
/plugin marketplace add mksglu/context-mode
/plugin install context-mode@context-mode
# 重启后验证
/context-mode:ctx-doctor
Gemini CLI / VS Code Copilot / JetBrains / Cursor / OpenCode / KiloCode / OpenClaw / Codex CLI / Pi Agent……15 个平台全支持，安装路径各有不同。
真实命令
// Before: 47 × Read() = 700 KB. After: 1 × ctx_execute() = 3.6 KB.
ctx_execute("javascript", `
const files = fs.readdirSync('src').filter(f => f.endsWith('.ts'));
files.forEach(f => console.log(f + ': ' + fs.readFileSync('src/'+f,'utf8').split('\n').length + ' lines'));
`);
# 终端跑诊断
context-mode doctor
context-mode upgrade
context-mode insight # 打开个人分析面板
基准数据
| 场景 | 原始大小 | 上下文大小 | 节省 |
|---|---|---|---|
| Playwright 截图 | 56.2 KB | 299 B | 99% |
| GitHub Issues ×20 | 58.9 KB | 1.1 KB | 98% |
| 访问日志 500 条 | 45.1 KB | 155 B | 100% |
| 分析 CSV 500行 | 85.5 KB | 222 B | 100% |
| Git 日志 153 commits | 11.6 KB | 107 B | 99% |
| 全会话总计 | 315 KB | 5.4 KB | 98% |
会话时间从 ~30 分钟拉到 ~3 小时。
隐私
什么都不传出去。无遥测、无云同步、无用量追踪。SQLite 全在本地，用完即焚。
已获 Hacker News #1（570+ 分），Microsoft / Google / Meta / Amazon / NVIDIA / ByteDance / Stripe 等团队都在用。
🔥 Context Mode: 14.6k Stars MCP Server for Context Optimization — 98% Token Savings Across 15 Platforms
Context Mode is an MCP server that solves one thing: keeping your AI coding agent's context window under control. A Playwright snapshot costs 56 KB. Twenty GitHub issues cost 59 KB. One access log — 45 KB. After 30 minutes, 40% of your context is gone. Context Mode's sandboxed tools compress 315 KB of raw output down to 5.4 KB — a 98% reduction.
Not filtering output — preventing raw data from ever entering the context window. Data is processed in isolated subprocesses; only summaries reach the model.
Four-Sided Solution
2. Session Continuity — Every event tracked in SQLite with FTS5 indexing. After compaction, BM25 search retrieves relevant state — model picks up where it left off
3. Think in Code — Write a script instead of reading 50 files. One ctx_execute() replaces ten tool calls, saves 100x context
4. No prose-style enforcement — Routes data away from context without telling the model how to talk
Installation (Claude Code)
/plugin marketplace add mksglu/context-mode
/plugin install context-mode@context-mode
# Verify
/context-mode:ctx-doctor
Supports 15 platforms: Claude Code, Gemini CLI, VS Code Copilot, JetBrains Copilot, Cursor, OpenCode, KiloCode, OpenClaw, Codex CLI, Antigravity, Kiro, Zed, Pi, OMP, Qwen Code.
Benchmarks
| Scenario | Raw | Context | Saved |
|---|---|---|---|
| Playwright snapshot | 56.2 KB | 299 B | 99% |
| GitHub Issues (20) | 58.9 KB | 1.1 KB | 98% |
| Access log (500 req) | 45.1 KB | 155 B | 100% |
| Analytics CSV (500 rows) | 85.5 KB | 222 B | 100% |
| Git log (153 commits) | 11.6 KB | 107 B | 99% |
| Full session total | 315 KB | 5.4 KB | 98% |
Session time extends from ~30 min to ~3 hours.
Privacy
Nothing leaves your machine. No telemetry, no cloud sync, no account. SQLite lives in your home directory, dies when you're done.
Hacker News #1 (570+ points). Used across teams at Microsoft, Google, Meta, Amazon, NVIDIA, ByteDance, Stripe.
GitHub: mksglu/context-mode | Install: npm install -g context-mode