ZoomInfo has put ZoomInfo MCP and the ZoomInfo API inside Snowflake, and the announcement is specific about how the work splits. Business users get a conversational layer. Snowflake CoWork and Cortex Agents call ZoomInfo’s MCP server as an external connector, with 14 tools in three groups: Finding (Search Companies, Search Contacts, Find Recommended Contacts, Search Intent, Search Scoops), Enriching (Enrich Companies, Contacts, Intent, Scoops, News) and Researching (Account Research, Contact Research). Data engineers get the API, called from Snowflake Tasks, Streams, Snowpark or external functions. “Bulk enrichment handles up to 10,000 records per request and 200,000 records per query.” Both sit alongside the Data Cubes ZoomInfo already ships to Snowflake Marketplace.
Some of these figures are claims and some are specifications. The GTM Context Graph’s “100M companies, 500M contacts, and billions of signals” is ZoomInfo’s description of its own database and can’t be checked from outside. The operational numbers can: 25 requests per second on the Standard API, 33 on the Scaling API, OAuth 2.0, the user’s existing role-based permissions carried into Snowflake, and audit logging on every call. The permissions point is the one to keep. A rep in CoWork “only see[s] contact data their ZoomInfo seat already entitles them to,” so the MCP connector doesn’t quietly widen access.
The catalog shows how much larger the API estate is than the MCP’s 14 tools. The ZoomInfo provider page lists 42 API pages. The story’s Enrich, Lookup, Bulk and Monitoring endpoints correspond to the ZoomInfo Company Enrich API, ZoomInfo Contact Enrich API, ZoomInfo Lookup API, ZoomInfo Bulk API and ZoomInfo Monitoring API/Webhooks API. The tiers it names include the ZoomInfo Company Master Data Enrich API and the ZoomInfo OrgChart Enrich API, and the MCP’s search and scoop tools sit over the ZoomInfo Intent Search API and the ZoomInfo Scoop Search API. The agentic-access profile counts 199 operations: 83 act on data and 5 are flagged human-in-the-loop.
ZoomInfo has a Kin Score of 87.2, in the exemplar band. Developer ergonomics at 92.9, access clarity at 92.1 and operational transparency at 92.1 carry it, and contract governance at 31.8 is its weak facet. Agent Readiness is 64.3, with the MCP server, rate-limit signal, delegated identity, protected resource metadata and dynamic client registration all lit. Those match what the story describes. Two dimensions are unlit, and both matter for warehouse pipelines: idempotency and pagination. A Stream that re-fires a Task can resubmit a 10,000-record enrichment job, and the published contract doesn’t say whether doing that twice is safe.