Today, Nimble is launching two new skills for AI agents: nimble-web-expert and nimble-agent-builder.

Install them into Claude, Cursor, Vercel, or any agentic tool of your choice, and your agent can now access live web data — on demand, at scale, and without you writing a single line of scraping code.

nimble-web-expert — Get web data, right now

nimble-web-expert gives your agent five commands for retrieving live data from anywhere on the web:

• nimble agent — run a pre-built extraction agent for a known site (Amazon, eBay, Shopify stores, and more)
• nimble extract — fetch and parse any URL, handling JS rendering and bot protection automatically
• nimble search — search the live web across 8 focus modes and get structured results back
• nimble map — discover all URLs on a site or sitemap before scraping
• nimble crawl — bulk-archive a section of a site asynchronously

In practice, this looks like asking Claude: "What are the top 10 best-selling running shoes on Amazon right now, with prices and ratings?" — and getting a clean, structured answer in under a minute. No API key for a specific retailer. No scraper to maintain. Just data.

nimble-agent-builder — Build extraction workflows that scale

nimble-web-expert is built for the moment. nimble-agent-builder is built for the long run.

When you have a recurring data need, describe what you want and the skill takes it from there. First, it searches the Nimble agent library for an existing agent that fits. If one exists, it shows you the schema, runs it with your parameters, and returns results — no build step required.

If no agent exists, or the existing one needs changes, the skill generates or refines one via MCP: it validates the output against real URLs, loops and refines if anything looks off, and only publishes once the output is clean. From that point, the agent is in your library and callable by name.

nimble agent CLI — find and run existing agents:

• nimble agent list — search the agent library by domain
• nimble agent get — inspect an agent's schema and input parameters
• nimble agent run — execute an agent with your inputs

nimble MCP — create, refine, and publish agents:

• nimble_agents_generate — build a new agent for a target site
• nimble_agents_update_from_agent — refine or clone an existing agent
• nimble_agents_publish — publish the validated agent to the library

Ask Cursor: "I need to extract product name, price, rating, and review count from any Shopify store product page." If a Shopify agent already exists, you get data immediately. If not, one gets built, validated, and published — ready for every run after that.

Need to extend it later? "The Shopify agent is missing inventory status — add it." The skill refines the existing agent rather than starting from scratch.

Nimble Skills aren't just tools — they're experts

Nimble Skills come loaded with the knowledge to use those commands well — which one to reach for in which situation, how to recover from failures, and how to best apply technical parameters. Your agent doesn't just gain new capabilities, it gains the expertise to use them effectively.

The skills are also connected to the Nimble docs via MCP, so they can look up new information in real time if they need it — staying current without manual updates.

And they get sharper over time. Every result you approve or correct is a signal. The skills learn your preferences and quietly optimize around them. The more you use them, the better they get.

How they work together

When Claude or Cursor has both skills installed, you never have to decide which one to call. The agent reads your intent — get data now vs. build something durable — and routes automatically.

Once an agent is built and published via nimble-agent-builder, it's immediately available inside nimble-web-expert via nimble agent run — so your reusable extraction workflows are always one command away.

Both skills share a feedback loop: every result you approve or correct improves the skill's routing and output quality over time.

What this makes possible

If you can imagine a data-driven app, you can now build it.

An agent that tracks competitor prices across every major retailer. A pipeline that monitors job postings across the web and alerts you to new openings the moment they appear. A research tool that pulls live product reviews, structures them, and feeds them directly into your workflow. Whatever you can describe, your agent can now go get.

The hardest part of building with data has always been the data itself — getting it, keeping it fresh, scaling it. That's where ideas used to stall. Not anymore.

AI agents bring the logic and the know-how. Nimble brings the data. Together, there's nothing left standing between you and whatever you want to build.

Get started

Prerequisites

npm install -g @nimble-way/nimble-cli

export NIMBLE_API_KEY="your-api-key"

Get your API key from Nimble Account Settings.

Install

For Claude Code:

claude plugin marketplace add Nimbleway/agent-skills &&

claude plugin install nimble@nimble-plugin-marketplace

For Cursor:

npx skills add Nimbleway/agent-skills -a cursor

For Vercel and other agents:

npx skills add Nimbleway/agent-skills

Full documentation at Nimble Agent Skills Guide.

‍