The free toolkit for agent builders.
Free, instant and private utilities for LLM apps, tool calling and MCP. No signup, nothing leaves your browser, and your agents can use them too.
Works with Claude Desktop, Cursor and any other MCP client.
Connect your AI agent in one paste
{
"mcpServers": {
"agenttoolbox": {
"url": "…/api/mcp"
}
}
}
{
"input_tokens": 1284,
"claude-opus-5-5": 0.0051,
"gpt-6.1-sol": 0.0026
}
✓ 4 tools available
Know the cost
before you ship.
Paste any prompt and see token counts and per-call and monthly cost across the top-ranked models from Anthropic, OpenAI, Google, xAI and DeepSeek, side by side.
Open Token & Cost Estimator| Model | Per call | Per month |
|---|---|---|
| Claude Opus 5.5 | $0.01400 | $14.00 |
| Gemini 4 Argon | $0.00700 | $7.00 |
| GPT-6.1 Sol | $0.00700 | $7.00 |
| DeepSeek V4.1 Flash | $0.00090 | $0.90 |
One schema.
Every format.
Turn a JSON Schema into ready-to-paste tool definitions for OpenAI, Anthropic and the Model Context Protocol. Validated as you type.
Open JSON Schema → Tool Definition{
"name": "get_weather",
"description": "Get current weather",
"input_schema": {
"type": "object",
"properties": {
"city": { "type": "string" }
},
"required": ["city"]
}
}Prompts, without
the guesswork.
Catch hedging, conflicting rules, leaked secrets, missing output formats and more, before your agent does something surprising.
Open System Prompt Linter⚠ Hedging language weakens instructions
✕ Possible API key found in prompt
⚠ No output format specified
✓ Role defined
Wire up MCP in seconds.
Generate mcpServers JSON for Claude Desktop, Cursor and others.
Open MCP Server Config Generator// claude_desktop_config.json
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "~/projects"]
},
"github": {
"url": "https://api.githubcopilot.com/mcp/"
}
}
}Schedules you can read.
Explain cron schedules for agent jobs and preview next runs.
Open Cron Expression Explainer*/15 9-17 * * 1-5
✓ Every 15 min, 09:00–17:45, Mon–Fri
Next runs
Mon 09:00 · 09:15 · 09:30
⚠ Server TZ is UTC, not local
Parse model output reliably.
Test regexes that extract structured data from LLM output.
Open Regex Tester for Agent Outputs/"amount":\s*(\d+\.\d2)/g
✓ 3 matches
$1 = 19.99
$1 = 4.50
$1 = 120.00
✕ line 7: "amount": "12" (string)
What it is Free browser tools and a hand-curated directory for building AI agents, also available to agents over a REST API and an MCP server.
Who it's for Developers working with LLM APIs, tool calling and MCP.
How it compares No signup, and the web tools run in your browser, so what you paste stays on your device.
Latest posts
Agent Skills: how to write a SKILL.md your agent actually loads
Agent Skills package instructions, scripts and reference files so an agent only pulls them in when a task calls for them. Here is how the loading works, how to write a description that triggers, and the portability and security traps to check before you ship one.
Read article →MCP extensions: how the opt-in layer works and which ones are worth adopting
The 2026-07-28 MCP revision made extensions an official part of the protocol and moved tasks out of core. Here is how extensions are negotiated, what the lifecycle stages promise, and how to decide whether your server should use Apps, Tasks, Skills or the auth extensions.
Read article →Running an agent on a local model: the tool-calling settings that quietly break it
Ollama, vLLM and llama.cpp all speak OpenAI-style tool calling, but each has a default that can wreck an agent loop, from a 4k context window to unconstrained arguments. Here is what to change before you blame the model.
Read article →Model routing for agents: when sending easy work to a cheap model actually cuts the bill
Routers, cascades and mid-run model switches, priced on real November 2026 rates, including the prompt-cache penalty that makes switching models inside a loop more expensive than it looks.
Read article →Field notes for people building AI agents. RSS
A directory worth bookmarking
Frameworks, MCP servers, sandboxes, observability and more, hand-picked and verified.
Firecrawl
Turn websites into LLM-ready markdown.
LangChain
Framework for building LLM apps and agents (Python/JS).
LangGraph
Stateful, graph-based agent orchestration.
LlamaIndex
Data framework for RAG and agents.
CrewAI
Multi-agent role-based orchestration.
AutoGen
Microsoft's multi-agent conversation framework.
OpenAI Agents SDK
Lightweight SDK for multi-agent workflows from OpenAI.
Vercel AI SDK
TypeScript toolkit for AI apps, tool calling and agents.
Your agents can use every tool.
A JSON API, an OpenAPI spec, llms.txt and a hosted MCP server. Point your agent at it and go.
curl -X POST /api/tools/estimate_tokens \
-d '{"text":"You are a helpful agent…"}'Developer docs Private by design
Web tools run 100% client-side. Your prompts never touch our servers.
Instant
No logins, no loading spinners. Open a tool and you're working.
Frequently asked questions
What is Agent Toolbox?
Agent Toolbox is a small, independent site with free tools for AI agent builders, such as a token cost estimator, a tool-schema converter and a prompt linter, plus a curated directory of agent frameworks and infrastructure.
Is Agent Toolbox free?
Yes. Every web tool is free, with no account. A paid Pro plan only applies to heavier use of the API and MCP server.
Can AI agents use Agent Toolbox?
Yes. The tools are available as a JSON API, an OpenAPI spec, llms.txt and a hosted MCP server. See the developer docs.
Does Agent Toolbox store my prompts?
No. The web tools run in your browser, so text you paste is processed on your device. See the privacy policy.
The daily agent toolkit.
New tools, MCP servers and frameworks, in one short email every morning. Free, no spam. Sign up and get the free AI agent cost cheat sheet (PDF).