Ask Claude About Your Partner Pipeline: The CinnaLab AI Connector

Partner managers do not avoid their PRM because it lacks features. They avoid it because the question in their head does not match the screens in front of them.

The question is "which partners registered something last quarter and have gone quiet since?" The screens are a partner list, a lead list, a deal list, three filters and an export. So the question gets asked in a chat instead, somebody answers from memory, and the data in the PRM slowly stops being the truth.

The CinnaLab AI connector is our answer to that gap. It lets an AI assistant such as Claude read your live CinnaLab workspace and, if you allow it, take a small set of actions in it. This post covers what it does, what it deliberately does not do, and how to set it up.

What it is

CinnaLab.io has a built-in MCP server. MCP, the Model Context Protocol, is an open standard for connecting AI assistants to the systems where data actually lives. Instead of pasting a CSV into a chat window and hoping the model reads it correctly, the assistant queries the source and works with the current record.

That matters more for a PRM than for most systems, because partner data goes stale quickly. A deal moves stage, a partner finishes onboarding, a lead is registered. An export is a photograph of a moment that has already passed.

Because MCP is an open standard, you are not tied to one AI vendor. This post uses Claude as the example. Other MCP clients, such as ChatGPT, Cursor or an agent you built yourself, connect to the same server.

What you can ask

Once connected, the assistant can read your partners, leads, deals, your partner program setup (including tiers and commission structures) and your program KPIs. With write access granted, it can also create and update leads, register deals, move deals between stages and invite new partners.

In practice that turns weekly reporting work into a conversation. The questions I hear from partner managers most often look like this:

The answers come from your workspace at the moment you ask. You can follow up, narrow it down, or ask the assistant to draft the email to the partner who has gone quiet. The drafting happens in Claude. The data stays in CinnaLab.

What it deliberately cannot do

Your AI assistant cannot reconfigure your partner program. Tiers, commission rates and program rules stay human-only.

Reading a commission rate is useful. Changing one is a commercial decision with contractual consequences for every partner it touches, and it should never happen because an instruction was worded loosely. So the connector simply has no tool for it. There is also no delete tool of any kind. This is enforced by the server, not by a line in a prompt, so no clever instruction moves it.

Security, in plain terms

Setting it up

There are two ways to connect, depending on the client.

With a token. In CinnaLab, open Settings, then AI Connectors, and generate a token. Choose its scopes and expiry, and copy it: this is the only time it is shown. Then add CinnaLab as a connector in your MCP client and paste the token. This is how Cursor, Claude Code and most custom agents connect.

With sign-in and consent. For Claude's custom connectors, you add CinnaLab as a connector and sign in with your CinnaLab account. A consent screen shows which workspace is connecting and exactly what is being requested, read-only by default, before anything is granted. Nothing is shared until you click Allow.

The AI connector is included in every paid CinnaLab plan: Starter, Growth and Scale, including paid-plan trials. It is not part of the Freemium plan; Freemium workspaces can see it in Settings and upgrade to switch it on.

Claude and Teams: the same rule in two places

We also built a CinnaLab app for Microsoft Teams, included in the same paid plans, where your team can @mention CinnaLab and ask the same kind of question inside a channel. The two are built on one rule: the assistant never answers as itself. It answers as the person, or the token, that is asking, with exactly that access. In Teams a partner only ever sees their own data, and a connector token only ever reaches its own workspace.

I see Teams as the place for the quick question in the middle of the day, and Claude as the place for the longer analysis: the quarterly review, the "why is this partner stalling" investigation, the draft you want to send afterwards. More on the Teams side in Run your partner program from Microsoft Teams.

The honest caveat

A better way to ask questions does not make a weak partner program strong. If your partners are not producing, the connector will tell you faster and change nothing else. What it removes is the reporting work as the reason nobody looks.

You can read the full connector details on the AI connector page and compare plans on the pricing page.

-- Cédric

Related Articles