Three layers, not one
It is worth separating them, because the failure modes differ. Native connectors are maintained integrations with defined behavior. The Zapier catalog provides breadth for the long tail. Model Context Protocol servers give agents governed access to tools at runtime, which is a capability question rather than a data-movement question.
- ✦Native connectors — 33 maintained integrations across email, communication, advertising and finance tooling, with defined sync behavior and failure handling.
- ✦Zapier — one of those 33 native connectors, and through it 9,000+ additional apps for the long tail. Those apps are Zapier's ecosystem, not Ivorycom integrations.
- ✦MCP servers — runtime tool access for agents, granted per workspace and policy-checked on every call.
A connection should improve a workflow
The useful question about any integration is which operating path it repairs. Syncing records between two systems that disagree about identity produces two wrong systems faster. The things worth checking are whether context survives the transfer, whether ownership stays visible, and what happens when the far side fails.
- ✦Does the context that justified the record travel with it, or only the fields?
- ✦Is ownership unambiguous after the sync, or does the record arrive unassigned?
- ✦What happens on failure — does it retry, surface, or fail silently?
- ✦Which system is authoritative when the two disagree?
Agent tool access is an authorization decision
An MCP connection is not a data pipe; it is a grant of capability to something that decides for itself when to use it. Ivorycom treats it accordingly. Access is granted per workspace, an agent can only invoke skills it has been given, and every invocation is authorized against your policy before it runs — with a failed or unreachable check denying the action rather than allowing it through.
- ✦Capability granted explicitly per workspace and per agent.
- ✦Every tool call authorized before execution, never after.
- ✦A policy check that cannot complete denies the action.
- ✦Invocations written to an append-only audit record.
How agent skills and channels workPolicy and audit
Run AI on your own keys
Workspaces can supply their own provider credentials for Anthropic, OpenAI, Google or Grok, so model usage runs on your account under your commercial terms. The model tier is selected per workspace from the provider's live model list rather than pinned to an identifier we chose, which means a new model becomes available to you without waiting for us to ship a release.
- ✦Bring your own keys, held per tenant.
- ✦Model selection from the provider's live list rather than a hard-coded identifier.
- ✦Provider choice per workspace rather than a platform-wide default.
Integrations and MCP in the product
Migration is the first integration
For most teams the connection that matters first is the one that brings the existing CRM across. Guided migration reads contacts, accounts, deals, custom fields and history from HubSpot, Salesforce, Pipedrive and Zoho, maps fields for review, and imports with validation, deduplication and a report.
- ✦Objects and custom fields read directly from the source system.
- ✦Mapping proposed automatically, confirmed by you.
- ✦Validation, deduplication and a migration report on completion.
Guided CRM migration
Frequently asked questions
How many integrations does Ivorycom have?
Thirty-three native integrations, and the Zapier connector is one of them. Through the Zapier ecosystem you can reach 9,000+ additional apps — that figure is Zapier's, not ours, and those apps are not Ivorycom integrations. The distinction matters because a native connector is maintained by us with defined sync and failure behavior, while a Zapier path is not. Natively the categories are email, communication, advertising, finance and productivity tooling; MCP servers give agents governed access to tools beyond both. Ask about the systems on your list.
What is MCP and why does it matter here?
Model Context Protocol is a standard way to expose tools to AI agents. It matters because it turns integration into an authorization question: the agent decides when to use a tool, so the grant has to be explicit and every invocation has to be checked against your policy.
Can we use our own AI provider keys?
Yes. Workspaces can supply their own Anthropic, OpenAI, Google or Grok credentials so model usage runs on your account, and the model tier is chosen per workspace from the provider's live list.
What stops an agent from using a connected tool badly?
It can only invoke skills it has been granted, each invocation is authorized against your workspace policy before it runs, an unavailable policy check denies rather than permits, and every call is written to an append-only audit record.
Can we migrate from our current CRM?
Yes — HubSpot, Salesforce, Pipedrive and Zoho are supported by guided migration, which carries contacts, accounts, deals, custom fields and history with validation, deduplication and a report.
What happens when an external system fails?
Failure handling is part of what makes an integration worth having. The right questions to ask of any connector are whether it retries, whether the failure surfaces to an operator, and which system is authoritative when the two disagree.
Ivorycom