工具说明为英文。

令牌效率

降低令牌成本并压缩上下文的工具。

用途

活跃度

排序方式

96

工具排名

Ranked by known GitHub stars, highest first; archived repositories come after maintained repositories. 原始值保留各自的数据窗口;计算变化需要 7 天内至少两次测量。

  1. 1
    休眠

    Analyze your MCP setup: token costs, grades, duplicates, and optimization tips

    mcp实测增长打开来源 ↗

    安装 claude mcp add mcp-checkup -- npx mcp-checkup

    1GitHub 星标稳定
  2. 2

    MCP
    活跃

    Context optimization for LLM agents: tool registry, result masking, budgeting, compaction.

    mcp实测增长打开来源 ↗

    安装 claude mcp add distil -- npx @munhq/distil

    1GitHub 星标稳定
  3. 3

    其他
    休眠

    Agent-to-Agent coordination that doesn't waste your context window. Token-efficient protocol with progressive discovery, zero-schema invocation, gRPC transport, and task lifecycle management.

    github实测增长打开来源 ↗

    1GitHub 星标稳定
  4. 4
    活跃

    Reversible context compression for AI agents: cut token usage up to 94%, originals retrievable

    mcp实测增长打开来源 ↗

    安装 claude mcp add token_optimizer -- uvx slimctx

    1GitHub 星标稳定
  5. 5
    活跃

    Cut AI agent token costs 5-15x — routes only relevant code symbols instead of full files.

    mcp实测增长打开来源 ↗

    安装 claude mcp add agent-booster -- uvx agent-booster

    1GitHub 星标稳定
  6. 6
    活跃

    A skill that shrink noisy log/output 60-90% before read — filter, group, dedupe, truncate. Cut token, cut cost, keep signal. By Dhayalan (aka Joy).

    github实测增长打开来源 ↗

    安装 /plugin marketplace add Da-ya7/PromptCompressor-skill

    1GitHub 星标稳定
  7. 7

    技能
    休眠

    Agent context gate for Codex, Claude Code, Copilot, MCP, Cursor, Cline, Gemini and Windsurf repos

    github实测增长打开来源 ↗

    安装 git clone https://github.com/grnbtqdbyx-create/contextforge ~/.claude/skills/contextforge

    1GitHub 星标稳定
  8. 8

    技能
    活跃

    Cut Claude Code token costs by delegating bulk LLM work to cheap models. 600+ models via NanoGPT, OpenRouter, Groq or local Ollama — batched, resumable, with real per-call cost tracking.

    github实测增长打开来源 ↗

    安装 /plugin marketplace add bicced/sidequest

    1GitHub 星标稳定
  9. 9
    活跃

    Context GC for LLM agents: offload large tool outputs and recall them to save tokens.

    mcp实测增长打开来源 ↗

    安装 claude mcp add lethe -- uvx lethe-llm-context

    0GitHub 星标稳定
  10. 10
    活跃

    Review and compact prompts, docs, and AI-agent skills to save tokens, preserving structure.

    mcp实测增长打开来源 ↗

    安装 claude mcp add compactprompt -- uvx compactprompt

    0GitHub 星标稳定
  11. 11
    活跃

    Verifies token/cost savings claimed by AI context-reduction proxies via MCP tools.

    mcp实测增长打开来源 ↗

    安装 claude mcp add tokentrust -- uvx tokentrust-cli

    0GitHub 星标稳定
  12. 12
    活跃

    Statically audits MCP tool surfaces for token cost, schema quality, and design issues.

    mcp实测增长打开来源 ↗

    需在 MCP 配置中声明的远程服务器 https://mcplint-web.vercel.app/api/mcp

    0GitHub 星标稳定
  13. 13
    活跃

    MCP server that cuts AI coding agent token usage via framework-aware context optimization

    mcp实测增长打开来源 ↗

    安装 claude mcp add ai-optimizer -- npx @ai-optimizer/core

    0GitHub 星标稳定
  14. 14

    技能
    休眠

    Drop-in Cursor rules to cut AI response tokens (no filler/hedging, preserve code + technical strings). Includes local, reproducible benchmarks.

    github实测增长打开来源 ↗

    安装 git clone https://github.com/marcomondelli/bonsai ~/.claude/skills/bonsai

    0GitHub 星标稳定
  15. 15
    未知最近提交:暂无数据

    AI agent token-cost telemetry + 429 prediction. Per-agent attribution, anomaly + routing + quota.

    mcp实测增长打开来源 ↗

    安装 claude mcp add openclaw-cost-tracker-mcp -- uvx openclaw-cost-tracker-mcp

    1297 天下载量稳定

学习与参考资源

Ranked by known GitHub stars, highest first; archived repositories come after maintained repositories. 这些资源可单独访问,不参与主要排名。

  1. 1
    活跃打开来源 ↗

    A cheat sheet that stops AI agents from scanning your whole repo and saves you tokens.

    github

    安装 git clone https://github.com/toshon-jennings/SUMMARY-md ~/.claude/skills/SUMMARY-md

    资源实测增长
    0GitHub 星标稳定

无公开指标的条目

Ranked by known GitHub stars, highest first; archived repositories come after maintained repositories. 原始值保留各自的数据窗口;计算变化需要 7 天内至少两次测量。

  1. 1

    MCP
    未知最近提交:暂无数据

    Pre-injection context compression for coding agents. Zero LLM calls, zero telemetry, offline-safe.

    mcp指标未公开打开来源 ↗

    无公开指标
  2. 2
    未知最近提交:暂无数据

    MCP server for Skim code transformation. Compresses code 60-95% for LLM context optimization.

    mcp指标未公开打开来源 ↗

    安装 claude mcp add skim-mcp-server -- npx skim-mcp-server

    无公开指标
  3. 3
    未知最近提交:暂无数据

    Cut LLM token costs: count tokens, estimate cost, slim prompts, and pick the cheapest capable model.

    mcp指标未公开打开来源 ↗

    安装 claude mcp add mcp-token-optimizer -- npx mcp-token-optimizer

    无公开指标
  4. 4
    未知最近提交:暂无数据

    LLM caching proxy (x402 USDC on Base) - exact + semantic cache. Free health.

    mcp指标未公开打开来源 ↗

    需在 MCP 配置中声明的远程服务器 https://cache.api.ainode.tech/mcp

    无公开指标
  5. 5
    未知最近提交:暂无数据

    Semantic context compression and cognitive memory layer for LLM swarms.

    mcp指标未公开打开来源 ↗

    需在 MCP 配置中声明的远程服务器 https://recallmax-mcp.vercel.app/api/mcp

    无公开指标