Delivery model

Nimble delivers web search results, extracted content, and research outputs wherever agents and data teams need them:

• To AI agents in real time via the MCP server, SDKs, and REST API

• Inside the data platform via native Snowflake and Databricks integrations that call Nimble from SQL

• To data stacks via async delivery to S3/GCS, or managed delivery

• As continuously refreshed datasets via scheduled Web Search Agents and Extraction Templates

Choose an integration path

All paths reach the same APIs. Pick by where your agent runs and how you want to authenticate. Most teams use more than one, for example a plugin for local development, MCP for a chat assistant, and the SDK in production, all with the same account.

Path	Best for	Auth	Start here
MCP server	Chat assistants and production agents	OAuth sign-in, or an API key	MCP Server Technical Spec
Plugin + skills	Coding agents with terminal access (Claude Code, Claude Cowork, Cursor, Codex, Cortex Code, Vercel Agent Skills CLI)	OAuth sign-in, or NIMBLE_API_KEY	Plugin Installation
SDK + CLI	Application code that needs full control	NIMBLE_API_KEY	Quickstart
Warehouse apps	Data teams working in SQL	API key held in the warehouse	Snowflake and Databricks

Using a coding agent? Paste Read and follow https://docs.nimbleway.com/agents.md into it and it installs Nimble itself. See Nimble Agent Setup.

What each path can reach

Product	Skills	MCP	SDK	CLI
Search, Extract, Extract Template, Map, Crawl, Web Search Agent	Yes	Yes	Yes	Yes
SERP, Media, Domain Knowledge, Jobs	No	No	Yes	Yes

Integrations and destinations

• AI assistants and coding agents: Claude (official plugin in the Anthropic & Partners directory, plus custom connectors), ChatGPT and Codex (native plugin), Cursor, VS Code and GitHub Copilot (via Azure MCP Center), Microsoft Copilot Studio, and Snowflake Cortex Code

• Agent frameworks: OpenAI (function calling and the Agents SDK), Anthropic SDK tool use, LangChain (langchain-nimble), LlamaIndex (llama-index-tools-nimble), Vercel AI SDK (@nimble-way/ai-sdk), Vercel eve (@nimble-way/eve), Google ADK, Mastra (@nimble-way/mastra), Cloudflare Agents SDK (@nimble-way/cloudflare-agents), and Hermes

• Built-in search provider: LiteLLM (search_provider="nimble", v1.98.0+), NVIDIA AI-Q Blueprint, OGX, and Omnigent

• MCP registries: Azure MCP Center, Databricks Marketplace, and Smithery

• Snowflake: Nimble Search and Nimble Agents Native Apps on the Snowflake Marketplace, NIMBLE_SEARCH/NIMBLE_EXTRACT UDFs for Cortex Agents, the NIMBLE_AGENT_RUN UDTF, App Builder Skills, and Cortex Code MCP

• Databricks: Marketplace install as a Unity Catalog MCP connection, Genie One, Genie Code skill, and nimble_search, nimble_extract, nimble_agent_run, nimble_agent_list, and nimble_agent_describe SQL table functions

• Other data platforms: BigQuery, S3, GCS, Azure (Microsoft Fabric, Azure Data Factory), AWS (S3, SageMaker, Redshift)

• Vector and BI: Pinecone, Power BI, Tableau

See Nimble in Snowflake and Databricks.

Implementation paths

API and SDK (self-serve)

Python (pip install nimble_python), TypeScript (npm install @nimble-way/nimble-js), and Go (go get github.com/Nimbleway/nimble-go) SDKs, the Nimble CLI (npm install -g @nimble-way/nimble-cli), and REST for Search, Extract, Extraction Templates, Map, Crawl, and Web Search Agents. Every SDK and the CLI read NIMBLE_API_KEY. Start free with 5,000 requests a month. Quickstart

MCP server and plugin (self-serve)

Connect any MCP host to https://mcp.nimbleway.com/mcp with OAuth sign-in, or install the Nimble plugin in a coding agent to get the MCP connection plus agent skills that teach the agent which Nimble product fits each job. MCP Server Technical Spec

Native warehouse integrations (self-serve)

Install from the Snowflake or Databricks Marketplace and call Nimble from SQL in notebooks, dbt models, and scheduled jobs.

Web Search Agents

Configure an agent's skill, goals, sources, effort, and output schema, or start from the agent gallery. Run it on demand or on a schedule.

Managed data services

Nimble's experts configure and operate the whole pipeline (collection, normalization, product matching, quality checks, and delivery) and stream data into your stack. Pricing is scoped with Nimble's sales team. See Web Scraping with Nimble.

How a request becomes usable output

1. Search finds candidate sources, guided by learned source preferences

2. Browse past search results into subpages, filters, and pagination

3. Extract clean content or typed fields

4. Validate with citations, confidence, schema enforcement, deduplication, and PII masking

5. Deliver to the agent or the destination

Onboarding and support

• 1:1 product onboarding and Technical Account Manager access for larger customers, arranged with Nimble's sales team

• Priority response with SLAs on custom arrangements

• Custom data normalization, enrichment, and product matching across sources

Enterprise options

Custom concurrent sessions, higher rate limits and burst capacity, throughput SLAs for peak events, bundled product discounts, and multi-year pricing protection. Contact sales.

Example pattern: enriching warehouse data

Customer systems remain the source of internal truth. Nimble adds external web context, such as pricing, availability, company facts, and news, so warehouse tables become continuously updated rather than static snapshots. With the native integrations, this runs as a SQL join against Web Search Agent output. A Nimble blog post on Snowflake and Upriver describes this pattern.