// category · 24 tools · 11 open source
AI Agent Tools & Protocols
The plumbing for agents: tool protocols, sandboxes, browser infrastructure, web data and search APIs.
Agents need more than a model and a loop. They need a standard way to reach tools, a safe place to run code, a browser to act on websites, and search APIs that return clean text. This category collects that infrastructure.
Protocols. The Model Context Protocol (MCP) is the common interface between agents and tools: you write a tool server once, and Claude Code, Cursor, the Vercel AI SDK, LangGraph and many others can use it. The current spec revision is dated 2026-07-28 and makes the protocol stateless. Learn the basics in how to build an MCP server in Python, see where it differs from a normal API in MCP server vs API, and compare it with agent to agent messaging in MCP vs A2A. Composio packages ready made integrations with hosted authentication for hundreds of apps.
Sandboxes. Code that a model writes should not run on your laptop or your production server. E2B and Daytona start isolated cloud machines in seconds through an SDK, so an agent can install packages, run tests and throw the machine away. Our E2B vs Daytona vs Modal guide compares startup time, limits and pricing.
Browsers. Browser Use, Stagehand and Playwright MCP let an agent click, type and read web pages, and Browserbase hosts the headless browsers they drive. See Browser Use vs Stagehand vs Playwright MCP for which to pick.
Search and scraping. Tavily and Exa return search results built for models, and Firecrawl turns whole pages or sites into clean Markdown. Many agents use one search API plus Firecrawl; the trade offs are in Tavily vs Exa vs Firecrawl.
When you add any of these to an agent, treat tool output as untrusted input: a web page or a tool description can carry instructions meant for the model. Limit what each tool can do, require approval for anything that spends money or deletes data, and pin versions so a compromised update cannot slip in. Official references: the MCP specification and the E2B documentation.
// guides
Guides to AI agent tools & protocols
Is E2B Sandbox Free? Hobby Credits and Limits
Is E2B sandbox free? The Hobby plan has no monthly fee and a one time $100 usage credit, then billing is per second. Here is what the credit buys and the limits.
Browser Use vs Stagehand vs Playwright MCP
Compare Browser Use, Stagehand, and Playwright MCP for AI browser agents: who owns the loop, languages, MCP fit, Browserbase hosting, and when to pick each.
Free MCP Server for Web Search: 7 Tested Picks
The best free MCP server for web search in 2026 is Exa's hosted endpoint, with SearXNG MCP for self hosting. We tested 7 options and their real free limits.
Vercel Agent Browser vs Playwright: CLI and MCP
Agent browser vs Playwright CLI and Playwright MCP, tested on one page: snapshot size, action output, MCP tool count, security controls and which to pick.
Best Self Hosted Tavily Alternatives (2026)
Self hosted Tavily alternatives for agents: Siftdog and Trawl drop in APIs, Firecrawl and crw for scrape jobs, plus when to stay on hosted Tavily.
Is Tavily API Free? Free Tier and Keyless Mode
Is Tavily API free? Yes, three ways: keyless calls, a free key with 1,000 credits a month, or a student plan. See what each covers, credit math and limits.
How to Start an AI Automation Agency
How to start an AI automation agency in 2026: pick a niche, build a demo, choose a stack, check n8n and Dify licenses, price the work and run agents safely.
Tavily vs Exa vs Firecrawl: Web Search for Agents
Tavily vs Exa vs Firecrawl for AI agent web access: what each does best, current free tiers and prices, cost per 1,000 searches, MCP servers and code examples.
E2B vs Daytona vs Modal: Agent Sandboxes Compared
E2B vs Daytona vs Modal for running AI agent code safely: isolation, startup, persistence, time limits, GPUs and SDKs compared, with code for each sandbox.
How to Use Tavily in LangChain: Step by Step
How to use Tavily in LangChain: install langchain-tavily, set your API key, add TavilySearch to create_agent, pin costly settings and handle errors the agent sees.
NVIDIA OpenShell: What It Is and How to Use It
NVIDIA OpenShell explained: how the open source agent sandbox enforces file, network and credential policy, how to install it, and how it compares to Docker.
Claude Managed Agents: Pricing and How It Works
Claude Managed Agents explained: what Anthropic hosts, the $0.08 session hour plus token pricing, self hosted sandboxes, budgets, limits and open source options.
Code Mode MCP Explained: Tool Calls as Code
Code Mode lets an AI agent write a short program that calls its MCP tools instead of calling them one by one. How it works, who supports it, and when to use it.
Chrome DevTools MCP: Setup, Tools and 1.13 Mocks
How to set up Chrome DevTools MCP in Claude Code, Codex, Cursor or Gemini CLI, what its 33 default tools do, the new 1.13 logpoints and mocks, and safe flags.
How to Use Firecrawl MCP: Setup, Tools and Limits
How to use Firecrawl MCP in 2026: add the hosted server with a key or sign in, run it locally over stdio, pick the right tool and know what keyless mode allows.
Is Tavily MCP Free? Yes, Within Credit Limits
Is Tavily MCP free? Yes. The server costs nothing extra. Keyless mode covers search and extract, and a free key covers all five tools with 1,000 credits a month.
Claude Code Email in User Agent: How to Fix
Claude Code put your account email in User Agent headers on web requests. What the 2.1.317 privacy mod covers, what it misses, and how to stop leaks today.
What Is MCP? The Model Context Protocol Explained
What is MCP? The Model Context Protocol explained: hosts, clients, servers, tools, resources, the stateless 2026 spec, security risks and a tested first run.
Cloudflare Security Audit Skill: How to Use It
How the Cloudflare security audit skill works, how to install it for Claude Code or Codex, what its six phases produce, and the sandbox rules it needs.
How to Use MCP Server in GitHub Copilot
How to use an MCP server in GitHub Copilot: the config file for VS Code, JetBrains, Visual Studio, Copilot CLI and the cloud agent, plus the GitHub MCP server.
How to Run an MCP Server Locally (2026 Guide)
How to run an MCP server locally with npx, uvx or Docker, test it with the MCP Inspector, then connect it to Claude Code, Cursor, VS Code or Codex safely.
How to Build an AI Agent from Scratch in Python
How to build an AI agent from scratch in Python: a tool loop, safe file tools, approvals, a step cap and memory in about 125 lines, on OpenAI or Ollama.
OpenCode 2.0: What's New and How to Upgrade
OpenCode 2.0 adds a shared background service, Code Mode for MCP and a new plugin API; 2.1 and 2.2 bring breaking SDK changes. What changed and how to upgrade.
AI Agents for Small Business: 8 Real Setups
AI agents for small business, made practical: 8 setups for shops, salons, restaurants, trades and freelancers, with the tools, steps and guardrails for each.