Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.nimbleway.com/llms.txt

Use this file to discover all available pages before exploring further.

Nimble brings the live web into Snowflake’s data and AI stack. Search, page extraction, site maps, crawls, and Web Search Agents sit alongside your warehouse tables, semantic models, and Cortex skills — so the AI you build inside Snowflake can ground its answers in current external signal, not just historical data.
Nimble web data tools running alongside Snowflake Cortex AI

​
Why web data belongs next to your warehouse

Snowflake’s AI surfaces — Cortex Code, Cortex Analyst, Cortex Agents, Streamlit — are great at orchestrating the data you already own. But the world your data describes lives outside the warehouse, and most of the questions stakeholders actually ask need both. Nimble closes that gap without leaving the Snowflake ecosystem:
  • Enrich tables in place. Pull current company metadata, product details, or pricing for the rows in a Snowflake table — and write the results straight back as a staging table.
  • Validate semantic models against the real world. Check whether the entities a Cortex Analyst semantic model points at still exist, still resolve, still match.
  • Ground Streamlit dashboards in live signal. Combine warehouse aggregates with current web context — competitor pricing, news mentions, regulatory updates — in the same view.
  • Build Cortex Agents that need the web. Give an agent the same Nimble tools you’d use yourself, so it can research, verify, and gather external context as part of its workflow.

​
Cortex Code

Snowflake’s native AI coding agent for the terminal — Cortex Code — is the first surface for connecting Nimble to Snowflake. One cortex mcp add command and an OAuth sign-in put Nimble’s full toolkit into every Cortex Code session.

​
What you get

  • Nimble’s web data toolkit in Cortex Code, exposed as mcp__nimble__* tools
  • One-command install — no JSON to edit by hand
  • OAuth 2.1 sign-in with PKCE — no API key to paste, manage, or rotate; tokens are stored in your OS keychain and refreshed automatically
  • Streamable HTTP transport at https://mcp.nimbleway.com/mcp

​
Prerequisites

​
Install Nimble MCP in Cortex Code

1

Add Nimble as an MCP server

Run a single command from your terminal:
cortex mcp add nimble https://mcp.nimbleway.com/mcp --transport http
Cortex Code writes the server to ~/.snowflake/cortex/mcp.json under the mcpServers key. No Authorization header is needed — Nimble’s MCP server advertises OAuth and Cortex Code handles the flow on first connection.
2

Sign in with OAuth

On the first connection, Cortex Code opens your system browser to authenticate. Approve the requested scopes — Cortex Code stores the access and refresh tokens in your OS keychain (mcp_oauth_nimble) and refreshes them automatically before expiry.
3

Verify the connection

List configured servers and trigger a connect:
cortex mcp list
cortex mcp start
cortex mcp start reports connected and failed counts and waits up to 300 seconds for each server to load. Once nimble shows as connected, its tools are namespaced as mcp__nimble__search, mcp__nimble__extract, mcp__nimble__map, mcp__nimble__crawl, and so on.
4

Use Nimble in a Cortex Code session

Start a session with cortex and ask the agent for anything that needs the live web. Cortex Code picks the right Nimble tool and runs it — no need to name the tool explicitly.
Cortex Code session calling mcp__nimble__search

​
Try these prompts

Paste any of these into a Cortex Code session — the agent picks the right Nimble tool and the right Snowflake skill, and chains them together.

1. Enrich a Snowflake table

“For each domain in ANALYTICS.CUSTOMERS.COMPANY_DOMAIN, use Nimble to extract the company’s current employee count and headquarters. Write results to ANALYTICS.STAGING.COMPANY_ENRICHMENT.”

2. Validate a semantic model

“Check each product URL in PRODUCT_CATALOG.URL with Nimble Extract — flag rows where the product is no longer available and write them to DATA_QUALITY.STALE_PRODUCTS.”

3. Build a Streamlit with live context

“Build a Streamlit app on SALES_MART.REVENUE that shows our top 10 accounts and uses Nimble Search to surface recent news for each one in a side panel.”

4. Prototype a Cortex Agent

“I’m scoping a competitive-research Cortex Agent. Use Nimble search and extract here so I can iterate on prompts and tool outputs, then turn the working chain into a Cortex Agent definition.”

​
Roll out Nimble across your organization

Cortex Code admins can enforce Nimble MCP for an entire Snowflake account through Cortex Code’s managed settings. Add Nimble to the administrator-enforced mcpServers block — and optionally to the URL allowlist — so every developer using Cortex Code gets the same connection without copy-pasting a command, and unsanctioned MCP servers are filtered out at merge time. 
{
  "mcpServers": {
    "nimble": {
      "type": "http",
      "url": "https://mcp.nimbleway.com/mcp"
    }
  }
}
Pair this with permissions.json to pre-approve safe Nimble tools (mcp__nimble__search, mcp__nimble__extract) and require confirmation on long-running ones (mcp__nimble__crawl).

​
Resources

Nimble MCP Server Docs

Full MCP Server setup for Cursor, Claude, and other clients

Cortex Code CLI

Snowflake’s docs for installing and using the Cortex Code CLI

Cortex Code MCP support

Transport types, configuration schema, and credential handling

Cortex Code managed settings

Enforce Nimble MCP across your Snowflake organization