Nimbleway

Nimble sells the retrieval layer between an AI agent and the live web. Its Web Search Agents are described as specialised per-domain retrievers with auditable 'Search Plans' recording what was searched, where and why.

• /search - real-time web search returning results, summaries and (in deep mode) page contents
• /agent - Web Search Agents that execute multi-step research workflows at a chosen compute effort level
• /extract - structured extraction from any public URL, with sync, async and batch modes
• /crawl - crawl entire sites or sections, with job listing and cancellation
• /map - enumerate site structure, optionally including the sitemap
• /dataset and Extract Templates - build, version and run reusable structured data feeds
• /monitor - watch for changes on the web (labelled beta)
• Browserless drivers (VX6 standard, VX8 JS rendering, VX10 JS + stealth) with auto-selection or manual override
• Hosted MCP server plus SDKs for Python, TypeScript and Go and a CLI

Engineering and data teams building agents, RAG pipelines or large structured web feeds - the published use cases are competitive pricing, digital shelf analytics, brand monitoring, alternative data for finance, influencer tracking and deep search for AI agents.

The self-serve API is genuinely pay-as-you-go, but the surrounding platform is not: the managed Data Services tiers on Nimble's own pricing page start at $2,500/month for Startup, billed annually, and rise to $7,000 (Scale) and $15,000 (Professional) - and it is those tiers, not the raw API, that carry MCP integration, custom agent ETL and data storage. Anyone whose plan is 'use the MCP server on a small budget' should verify what the pay-as-you-go tier actually includes before committing. Search is priced at $5 per 1,000 search inputs and extraction at $1.00 - $1.45 per 1,000 URLs depending on driver, so cost is driven by how much JS rendering and stealth your targets need - high-defence sites push you to the most expensive driver. The Agent API cost is published only as '$EFFORT (you choose compute)' with no figure attached, which makes budgeting for agentic research runs impossible from public pricing alone. And the free trial is 5,000 web pages, which is an evaluation allowance, not a usable free tier.

Setup assessment: 30m

Sign up, get an API key, install one of the SDKs and issue a Bearer-authenticated POST - the documented quickstart is a handful of lines. Getting to a production pipeline takes longer: choosing drivers, authoring or generating Extract Templates, and wiring a data destination such as S3, Snowflake or Databricks.

Prerequisites

• Nimble account and API key (free trial starts with 5,000 web pages)
• Bearer token auth on all API calls
• For streamed feeds: a destination bucket or warehouse (S3, GCS, Azure Data Lake, Snowflake or Databricks)

Authentication

Bearer token (`Authorization: Bearer <YOUR-API-KEY>`), declared as BearerAuth security on every operation in the OpenAPI spec. The MCP example uses `Authorization: Basic <api key>` against the SSE endpoint.

Base URL

https://sdk.nimbleway.com (v2 API; the OpenAPI spec also declares https://api.us.webit.live/api as the production server for the Fast SERP operation)

Rate limits

The OpenAPI spec declares a 429 'Rate Limit Exceeded' response on operations, and the enterprise tier advertises higher rate limits and burst capacity, but no numeric limit was published in what we retrieved.

Machine-readable API

OpenAPI specification linked

• POST /v2/extract, /v2/extract/async, /v2/extract/batch - structured extraction
• POST /v2/search - web search with focus, max_results and search_depth parameters
• POST /v2/crawl, GET /v2/crawl, GET /v2/crawl/{id}, DELETE /v2/crawl/{id} - crawl lifecycle
• POST /v2/map - site structure mapping with optional sitemap inclusion
• /v2/extract/templates - create, list, get, update, delete, version and run Extract Templates (realtime, async and batch)
• POST /v2/extract/templates/generations - generate an extract template
• GET /v2/domain-knowledge/driver - resolve the suggested driver for a URL or agent name
• Jobs API - create/get/update/delete jobs, trigger runs, list runs, cancel a run, list/preview/download run artifacts

• Numeric API rate limits - a 429 response is specified but no published figure was retrieved
• The per-effort cost of the Agent API - published only as '$EFFORT'
• Whether MCP access is available on the pay-as-you-go API tier or only on the paid Data Services tiers; the pricing page lists 'MCP integration' as a Data Services feature
• Official social profiles and any official vendor video