This feature is currently in closed-beta. If you are interested in participating in the beta program, please contact your Customer Success Manager (CSM).
This article explains the reasoning behind the design, and it is most useful when Claude does something you did not expect.
The sections cover why Cint MCP has only a connector, why an organization owner must add the connector, why the values you leave out of a request affect the result, which behaviors look like errors but are not, and what Cint can see when you use Cint MCP.
Why Cint MCP has only a connector
Earlier rounds of the beta had two parts: a connector and a plugin. The plugin carried the skills, which are the written guidance that tells Claude how to run a feasibility check, how to set quotas, and what to check before a launch. That guidance now comes from the Cint server itself.
When Claude connects, the server tells it how to work with the Exchange. Claude loads more detailed guidance from the server as a task needs it. You may see this in the conversation as a “load skill” tool call.
Because the guidance lives on the server, it changes when the server changes. Everyone gets the same version at the same time. There is no file to send or install, and an improvement reaches you without an owner approving anything or anyone signing in again.
The guidance also always matches the tools. A skill installed on your own machine can fall behind the server it describes. Guidance served by the server cannot.
An organization owner still adds the connector once, and each person still signs in with their own Cint account.
See Exchange: Cint MCP Install and Connect in Claude for more information on adding the connector.
Why an organization owner adds the connector
On Claude Team and Enterprise plans, only an organization owner can add a custom connector. This is a Claude requirement, not a Cint restriction, and Cint cannot change it.
Connector access also cannot be limited to part of an organization. After an owner adds the connector, it is available to everyone in the organization.
Adding the connector does not share anyone's login
The connector holds no credentials. After an owner adds it, each person signs in with their own Cint account. Claude can access only what that account can access, and nothing that belongs to a colleague.
Why values you leave out affect the result
Many fields in a feasibility check have a default value. If you leave a field out, Claude uses the default. A default can produce a confident, precise answer to a question you did not ask.
For this reason, Claude never assumes the two numbers that move a feasibility result most: the interview length and the incidence rate. Claude asks for them instead.
Claude also reports what a call actually sized and used, not the values you asked for. This line is the safeguard, and it works only if you read it.
Behaviors that look like errors but are not
Some Cint MCP behaviors look wrong at first but are expected. Find what you are seeing in the tables below.
Terminology and structure
What you see | Why it happens |
|---|---|
Claude says “target group” when you | The survey is your questionnaire, which lives on your own platform. The target group is the Exchange entity that fields it. It defines the market, qualifications, quotas, goal, price, supply, and dates. Claude uses the precise term so that statements like “three of your five target groups are live” are exact. You can keep saying “survey.” |
You asked for one survey in five markets | This is the correct structure: one questionnaire, five target groups, one project. Each market is priced, quota’d, and fielded separately. |
More qualifications than you asked for | A geographic qualification adds the qualifications that depend on it, usually postal code. The platform adds them to resolve the geography. |
Creating and launching
What you see | Why it happens |
|---|---|
Claude creates a draft when you asked it | Nothing goes into field without an explicit launch and an approval prompt. The draft is where you check the price and the goal. |
Claude shows a proposed quota split | When a qualification targets several values and has no quotas of its own, Claude proposes a split rather than inventing one. Approve the split, or give Claude your own numbers. |
Claude did not ask about supply | If you don’t specify supply, the target group uses your account’s default allocation. For most studies, this is the right choice. Ask Claude which allocation it used if you want to confirm. |
A new target group is named “Draft target | This is the default name when none is given. Tell Claude the name you want when you set up the target group. |
Claude says the goal counts “unknown,” | Fielding automation is on by default, as it is in the Exchange UI. The Exchange manages respondent flow to avoid overfilling quotas, and it decides what the goal counts. Automation does not reduce your fill. If you need the goal to count completes, ask for that when you set up the target group. This switches the automation off. |
Pricing and metrics
What you see | Why it happens |
|---|---|
A price you gave was ignored | The business unit uses a rate card, so the card sets the price of the target group. Claude should say so. If it does not, ask which pricing model the business unit uses. |
A price sits exactly on a round number | The price sits at the Exchange’s price floor or ceiling, which protect suppliers and buyers respectively. The estimate is sound. |
Metrics are blank or zero on a new | Counts are zero until respondents arrive, and rates are empty until there is data to calculate them from. This is not an error. |
What Cint can see
The connector authenticates as you and acts within your existing Cint permissions. It cannot reach anything your account can’t. Tool calls are logged for beta support. There is no separate credential to rotate or leak, because nothing stores one.
See the following for more information: