Claude Toolkit
28 curated agent templates for Claude Code, Cursor, Copilot, Gemini CLI, Windsurf, and MCP. Includes live Prompt Caching ROI Calculator, official October 2026 Anthropic pricing, and .ant.yaml / AGENTS.md generator.
⚡ Anthropic Prompt Caching & Token Cost ROI CalculatorVendor Verified
Calculate multi-turn agent token cost reduction using Anthropic's 90%–97.5% cache read discount (0.10x Haiku 5.5 · 0.05x Sonnet/Opus 5.5 · 0.025x Fable 5.1).
🛠️ Interactive Config BuilderUpdated
Claude Code & ant CLI
Anthropic's agentic CLI and Managed Agents GitOps harness (`ant apply`). Configured with 5-minute & 1-hour prompt caching (512-token min), deterministic lifecycle hooks, Agent Skills (`SKILL.md`), root `.mcp.json`, and Claude 5.5 / Fable 5.1 / Haiku 5.5 support.
Anthropic API & Prompt Caching
Production integration snippets using GA Anthropic Prompt Caching (`cache_control` 5m & 1h TTLs, 512-token min, up to 4 breakpoints), native 128K output, Adaptive Thinking (`output_config.effort`), and Conversation Compaction.
Cursor
AI-native code editor. Uses `.cursor/rules/*.mdc` (plain `.md` files in `.cursor/rules/` are ignored; legacy root `.cursorrules` is deprecated), open-standard `AGENTS.md`, and `.cursor/mcp.json` (40 active tool cap).
GitHub Copilot
GitHub's AI agent suite in VS Code & Cloud Coding Agent. Uses `.github/copilot-instructions.md` (repo-wide), `.github/instructions/*.instructions.md` (`applyTo` path-scoped), `.github/prompts/*.prompt.md` (slash runbooks), and `.vscode/mcp.json` (`servers`).
Gemini CLI & Antigravity
Google's agentic coding tools. Gemini CLI for terminal workflows and Antigravity for multi-agent IDE sessions. Configured via `GEMINI.md` (or `AGENTS.md`) and `.gemini/settings.json`.
Windsurf & Codex CLI
Windsurf (Cascade) modular `.windsurf/rules/*.md` (`trigger: always_on | glob | model_decision | manual`, legacy `.windsurfrules` <6,000 chars) and OpenAI Codex CLI hierarchical `AGENTS.md` steering.
Roo Code & Cline
Autonomous coding agent extensions for VS Code. Uses .clinerules or .roomode/ rules for custom agent personas (Architect, Code, Debugger, Security).
Zed AI Editor
High-performance GPU-accelerated code editor with native Anthropic Claude 5.5 integration, MCP servers, and custom prompt presets.
MCP Server Ecosystem
Production Model Context Protocol (MCP) configurations (`stdio` & Stateless Streamable HTTP) for Claude Desktop, Claude Code (`.mcp.json`), Cursor (`.cursor/mcp.json`), and VS Code (`.vscode/mcp.json`) — including Project SPINE and Project ChronoFact.
📊 Cross-Tool Cheat Sheet
| Feature | Claude Code (ant CLI) | Cursor | Messages & Agents API | Windsurf & Codex |
|---|---|---|---|---|
| Primary Config | .ant.yaml / CLAUDE.md | .cursor/rules/*.mdc | src/agents/client.ts | .windsurf/rules/*.md |
| Sandboxing Mode | local / docker | local / cloud VM | Cloud Native (Managed) | local sandbox |
| Scoped rules | .claude/skills/*/SKILL.md | globs in .mdc | container.skills | trigger: glob / AGENTS.md |
| API Headers | Native | Native | anthropic-version: 2026-10-21 | Native |
| Sub-agents | .claude/agents/*.md | Projects & Subagents | Claude Agent SDK | AGENTS.md |
| MCP config | .mcp.json / ~/.claude.json | .cursor/mcp.json | Streamable HTTP MCP | mcp_config.json |
2026 Enterprise AI Engineering Shift
AI coding workflows have evolved from basic inline completions into multi-agent autonomous engineering loops. Using Anthropic's GA cache_control: { type: "ephemeral" } (512-token minimum prefix on Claude 5.x) and the .ant.yaml GitOps CLI harness, teams retain 1-million-token contexts across entire monorepos without prohibitive token overhead.
Whether targeting Claude Sonnet 5.5 ($2/$10), Claude Opus 5.5 ($4/$20), Claude Fable 5.1 ($10/$50), or Claude Haiku 5.5 ($0.10/$0.50), 5-minute prompt caching achieves break-even ROI on the very second turn of any agent session (90%–97.5% cache read discount). Modular .cursor/rules/*.mdc, .claude/skills/*/SKILL.md, AGENTS.md, and CLAUDE.md files enforce deterministic architectural guardrails during high-speed refactoring sprints.
This toolkit provides 28 production-grade templates across 9 AI ecosystems. All templates are completely free, open-source, and run entirely client-side in your browser.
Frequently Asked Questions
What is the Claude Toolkit?
Claude Toolkit is a free, comprehensive developer suite featuring 28 copy-paste configuration templates (CLAUDE.md, .claude/settings.json hooks, SKILL.md, .mcp.json, .ant.yaml, .cursor/rules/*.mdc, AGENTS.md, .github/copilot-instructions.md, and .vscode/mcp.json), an interactive config generator, and an Anthropic Prompt Caching & Token Cost Calculator.
How does Anthropic Prompt Caching work and how much does it save?
Anthropic Prompt Caching caches static system instructions, tool schemas, and codebase prefixes (minimum 512 tokens across Claude 5.x, up to 4 breakpoints per request) for 5 minutes (1.25x write) or 1 hour (2.0x write). Subsequent cache hits cut input costs by 90% to 97.5%: $0.01/M on Haiku 5.5 (0.10x), $0.10/M on Sonnet 5.5 (0.05x), $0.20/M on Opus 5.5 (0.05x), and $0.25/M on Fable 5.1 (0.025x).
Which Claude models are supported?
Claude Toolkit natively supports Anthropic's October 2026 frontier lineup (all featuring 1M context and 128K max output): Claude Sonnet 5.5 ($2/$10), Claude Opus 5.5 ($4/$20), Claude Fable 5.1 ($10/$50 top reasoning), and Claude Haiku 5.5 ($0.10/$0.50 up to 100K tokens), alongside gated Claude Mythos 5, Series 5.0 baselines (Sonnet 5, Opus 5), and legacy Claude Haiku 4.5.
Does the toolkit support Conversation Compaction & Managed Agents?
Yes, Claude Toolkit includes verified TypeScript snippets for both Messages API Conversation Compaction modes — On-Demand summarization (compact-2026-09-04) and Threshold Auto-Compaction (compact-2026-01-12) — plus Anthropic Managed Agents auto permission policy and ant apply GitOps configurations.
Is the Claude Toolkit free to use?
Yes, 100% free with no sign-up or installation required. All configs, calculations, and templates run locally in your browser.