Agent Connect MCP
Agent Connect implements the Model Context Protocol, so the AI assistant your team already works in can reach your channel program directly. Ask a question in plain language and the assistant answers from your own data, under your own permissions, without anyone opening the admin console.
Works with the assistant your team already uses
The Model Context Protocol is an open standard for connecting AI assistants to the systems where work lives. Agent Connect implements it, so one connector serves every compliant client.
Claude
AnthropicAdd Unifyr from the connector settings in Claude Desktop or the Claude web app. Your channel team can pull partner records into the same conversation where they are already drafting and planning.
Microsoft Copilot
MicrosoftFor Microsoft-first organizations, the same connector puts partner records, deal status, and analytics answers into the assistant that already sits beside Outlook and Teams.
Gemini
GoogleGemini reaches Unifyr over the same protocol, so Google-first organizations get an identical capability set with no separate integration to maintain.
ChatGPT
OpenAIRegister Unifyr as a connector and ChatGPT can query your program and carry out admin work in the chat window your team keeps open all day.
Your AI strategy can change without your partner platform integration being rebuilt. Anything that implements the protocol connects the same way.
How the connection works
Three things make the connector safe to hand to a channel operations team rather than a developer.
Connect once
Authorize Unifyr in your assistant's connector settings. There is no middleware to deploy and no API client to write, because the protocol handles discovery of what the connector can do.
Inherit your permissions
The connector acts as the signed-in Unifyr user. Whatever a person cannot see or change in the console stays out of reach from the assistant, so your existing role model governs the session.
Approve every change
Anything that creates, sends, or updates a record is declared as a write, and your client asks for approval before it runs. Approved actions land in the same audit trail as work done by hand.
The intelligence comes from Unifyr IQ
Scoring and analytics answers run in the platform on your program's own history, not in the chat window.
Deal IQ
Ask whether a deal will close and the answer is Deal IQ's, scored against your own recent wins and losses, with the reasoning attached. Leads are scored the same way.
Analytics IQ
Questions about campaigns, revenue, or benchmarks go to Analytics IQ, which answers in business terms off the reporting layer rather than by reading a spreadsheet you pasted in.
Governance your security team will recognize
An AI connector to a partner platform raises the same questions as any other integration. Unifyr answers them with the controls that already govern the platform.
Reads and writes are separated
Retrieval runs freely. Capabilities that create records, change status, or send messages to real people are declared as writes, so the approval prompt in your client tells you what is about to happen.
No model training on your data
Program data reached through the connector is never used to train models, and the intelligence layer is isolated per customer with no intermixing between tenants.
Opt-in and audited
The connector is enabled per customer, so a security review can finish before anyone connects. Actions taken through it are logged alongside the rest of your platform activity.
The Model Context Protocol, usually shortened to MCP, is an open standard that describes how an AI assistant discovers and uses the capabilities of an external system. The assistant asks the connector what it can do, and the connector responds with a set of capabilities it can then invoke on your behalf.
Agent Connect is Unifyr's implementation of the protocol. The practical benefit is that the integration is written once. Because capabilities are exposed through the protocol rather than through a vendor-specific plugin format, the same connector works in Claude, ChatGPT, Gemini, Microsoft Copilot, and any other compliant client, including ones that do not exist yet.
Claude, ChatGPT, Gemini, and Microsoft Copilot are the clients most Unifyr customers use, and each connects through its standard connector settings. Custom agents built on your own stack connect the same way, since the protocol is what determines compatibility rather than a partnership with any individual AI vendor.
Roughly everything a channel administrator does on a normal day. Partner administration covers search, full record detail, creation, approval and rejection, status changes, and partner group suggestions. Deal registration covers pipeline search, stage and status transitions, ownership, value, comments, and contacts. Leads cover search, detail, and assignment either directly to a partner or through a distribution flow.
Beyond record management, the connector searches the collateral library either by its fields or by describing what you need, drafts and tests partner email campaigns, answers analytics questions in plain language through Analytics IQ, and scores deal and lead outcomes.
Unifyr. Ask whether a deal will close and Deal IQ scores it inside the platform against your own recent wins and losses, then the assistant reports what came back. Analytics answers work the same way through Analytics IQ, against the reporting layer rather than a guess. The assistant supplies the conversation; Unifyr IQ supplies the intelligence, and it is the same intelligence your team sees in the console and your partners see in the portal.
That distinction matters for trust. A general-purpose assistant reasoning over a pasted export can produce a confident number nobody can trace. Here the number has a source in your program. Unifyr IQ is where that work happens.
Both, and the difference is explicit. Read capabilities such as search, record detail, and analytics return information without touching anything. Write capabilities create partners, transition deals, assign leads, update collateral metadata, and send communications.
Every write is declared as one, which means your client surfaces an approval prompt describing the action before it executes. The writes that reach real people, such as approving a partner or sending a notification, are the ones worth reading carefully before you approve them. If your program prefers a read-only deployment, permissions on the connecting user are what enforce that.
The connector operates as the Unifyr user who authorized it and inherits that user's permissions. A channel manager who cannot approve partners in the console cannot approve them through an assistant either. Nothing about the protocol grants elevated access, so your existing role-based access controls remain the mechanism that decides what is possible.
No. Data reached through the connector is not used to train models, and Unifyr's intelligence layer is isolated per customer so your information is never intermixed with another customer's. Data is encrypted in transit and at rest, and Unifyr is SOC 2 compliant with role-based access controls. Our Responsible AI Policy sets out these commitments in full.
Record lookups such as partners, deals, leads, accounts, and collateral read your program directly. Analytics and reporting answers come from the reporting layer, which refreshes daily, so a figure in an analytics answer reflects the most recent refresh rather than the current minute. For pipeline questions where timing matters, ask about the deal records themselves.
Both audiences are addressed. Internal administrators get the program-wide capability set described on this page. Partners can connect their own assistant to the partner portal and work with the data that belongs to them, which is the same scoping the portal already applies to what a partner sees when they sign in.
Availability depends on your Unifyr plan, and the connector is enabled per customer, so the first step is a conversation with your Unifyr team about which surfaces you want available and which users should be able to authorize a client. Once it is enabled, connecting an assistant is a matter of authorizing Unifyr in that client's connector settings. Book a demo and we will walk through the capability set against your own program.