Skip to main content
The Nimble Agent Builder skill lets you create, test, refine, and publish reusable extraction workflows from your AI coding assistant. Describe the data you need from any website, and it finds an existing agent or creates a new one, then lets you iterate until the output is exactly right. Built on the Nimble MCP Server and the open-source Agent Skills standard for cross-platform agent compatibility.

​
Tools Overview

ToolDescription
nimble_agents_listBrowse available agents by keyword
nimble_agents_getGet agent details and input/output schema
nimble_agents_generateCreate custom agents using natural language
nimble_agents_updateRefine agent fields, parsing rules, or extraction logic
nimble_agents_runExecute agents and validate structured results
nimble_agents_publishSave finalized agents for repeated use

​
Quick Install

claude plugin marketplace add Nimbleway/agent-skills && \
claude plugin install nimble@nimble-plugin-marketplace
The agent builder requires the MCP server. See Prerequisites > MCP Server Connection below.

​
Prerequisites

​
1. Nimble API Key

Sign up and generate a key from your Account Settings > API Keys. Set the NIMBLE_API_KEY environment variable using your platform’s method:
Add to ~/.claude/settings.json:
{
  "env": {
    "NIMBLE_API_KEY": "your-api-key-here"
  }
}

​
2. MCP Server Connection

The skill requires a connection to the Nimble MCP server. After installing the skill, connect the server: 
claude mcp add --transport http nimble-mcp-server https://mcp.nimbleway.com/mcp \
  --header "Authorization: Bearer ${NIMBLE_API_KEY}"
See the Plugin Installation page for platform-specific setup options.

​
Quick Start

The skill activates automatically when you describe a recurring extraction need: 
"set up extraction for Amazon product pages"
  -> searches for existing agents, finds amazon_pdp, runs it, shows results

"create a reusable scraper for job listings on LinkedIn"
  -> generates a custom agent, lets you refine fields, publishes when ready

"build an agent for extracting hotel prices from Booking.com"
  -> creates agent from natural language, validates output, saves for reuse
Use this skill when you need a reusable workflow. For immediate, one-off data extraction, use the Nimble Web Expert instead.

​
How It Works

The skill follows a build-test-refine lifecycle:
  1. Search existing agents that match your target website
  2. Inspect the agent’s input/output schema
  3. Run the agent and validate structured results
  4. Generate a custom agent using natural language if no existing one fits
  5. Refine fields, extraction rules, or parsing until the output matches expectations
  6. Publish the final agent for repeated use across projects
Results are presented in clean markdown tables with numbered follow-up options. Each step is interactive. Refine until the output matches your requirements before publishing.

​
Usage Examples

​
Find and run an existing agent

"extract product details from this Amazon page: amazon.com/dp/B0DGHRT7PS"
The skill searches for a matching agent (amazon_pdp), shows you the schema, runs it, and returns structured product data.

​
Generate a custom agent

"build an agent that extracts job title, company, salary, and location from Indeed job listings"
The skill generates a new agent definition, runs it against a sample URL, and lets you refine the fields before publishing.

​
Refine and publish

"add a 'remote' field to the agent"
"the salary field is empty, try a different selector"
"looks good, publish it as indeed_jobs_v2"

​
Agent Builder vs Web Expert

agent-builderweb-expert
GoalReusable workflowImmediate data
SpeedSlower (build/test/refine cycle)Fast, direct
OutputPublished agentExtracted data
Use when”I’ll need this repeatedly""Get me this data now”

​
About Agent Skills

This skill follows the Agent Skills open-source standard, making it compatible with multiple AI agent platforms. Install using the Skills CLI, the standard package manager for the Agent Skills ecosystem.

​
Full Reference

For the complete skill documentation, examples, and API reference, see the SKILL.md on GitHub.