VegaDūtaIntegrations · Business systems
Connecting your CRM to an AI agent
Straight answer first: VegaDūta does not ship a native connector for Salesforce, HubSpot, Zoho, or any other specific CRM. What it does have is a general-purpose path that covers all of them — the Model Context Protocol connector registry. Any CRM that has an MCP server, or can host one, connects to your agents through it: browse the official MCP registry from the console, or register a custom server with encrypted credentials. Once connected, the CRM's operations become tools your agent can call in a conversation or workflow.
If you do not have a CRM yet, the platform has a small one built in: a leads list that workflows can write to. It is a place to track leads, not a replacement for a full CRM suite. We would rather tell you both things plainly than list CRM logos we cannot back up.
By the VegaDūta team · Last updated
How the MCP route works
MCP is the open protocol that lets AI systems call external tools. If your CRM vendor publishes an MCP server — a growing number do — or you host one against your CRM's API, VegaDūta connects to it like any other MCP server.
- Browse the official MCP registry directly from the console.
- Register a custom MCP server for a CRM not in the registry.
- Server credentials are stored encrypted.
- Connected servers pass admin review before agents can use them.
The built-in leads list
The admin console includes a leads list. A CRM lead node in a workflow creates or updates a lead, matched by email address, and each lead moves through stages: new, contacted, replied, qualified, proposal, won, or lost. The bulk email node sends personalised outreach with a recipient cap, a send-rate limit, and a suppression list. When a recipient replies, the reply is recorded on the lead and the lead moves to the replied stage; bounces are detected as well. Leads can be imported and exported as CSV.
What a CRM-connected agent can do
Whatever the MCP server exposes: look up a customer record mid-conversation on WhatsApp, create a lead from an Instagram DM, update a deal as a workflow step. The agent calls the CRM's tools the same way it calls any other tool, inside the same conversations and workflows.
Governance: review, guardrails, approvals
Third-party servers do not get a free pass. Registered MCP servers go through admin review, tool calls run under the platform's guardrails (PII, toxicity, jailbreak detection), and human approval gates can require sign-off before an agent writes to your CRM. Credentials stay encrypted at rest.
What to check before you connect
Two questions decide feasibility: does your CRM have an MCP server (check the registry from the console), and if not, can you host one against its API? If the answer to both is no, VegaDūta cannot connect to that CRM today — that is the honest boundary of this integration.
Frequently asked questions
Does VegaDūta integrate with Salesforce, HubSpot, or Zoho?
Not natively — there is no built-in connector for any specific outside CRM, and we say that plainly. If the CRM has or can host an MCP server, it connects through VegaDūta's MCP connector registry, and its operations become agent tools. A simple built-in leads list is available if you do not use an outside CRM.
How do I connect my CRM to an AI agent?
Through the Model Context Protocol: browse the official MCP registry from the VegaDūta console for an existing server, or register a custom MCP server pointed at your CRM's API. Credentials are stored encrypted, and the server passes admin review before agents can use it.
What can an agent do once my CRM is connected?
Whatever tools the MCP server exposes — typically reading records, creating leads, and updating fields. The agent calls them mid-conversation or as workflow steps, with guardrails and optional human approval gates on writes.
Is it safe to give an AI agent access to CRM data?
The platform is built to make it governable: encrypted credential storage, admin review of registered servers, guardrails including PII detection on tool traffic, and approval gates that put a human before sensitive actions. You decide which agents get CRM tools at all.
See it working in two minutes
The sandbox provisions a real tenant — describe an agent in one sentence and test it, no account, no card. Or browse ~90 industry workflow recipes to see what teams build.
