ChatGPT Team Setup Consultant: Deploy an MCP-Connected AI Assistant

A 12-person recruitment agency in Manchester pays for ChatGPT Team seats every month. Six of those seats get used for copy-pasting candidate summaries between Gmail, a spreadsheet and HubSpot, then re-typing the same boilerplate into Slack. Nobody has connected the assistant to anything. The subscription is active; the leverage is not.

That gap is what a chatgpt team setup consultant is usually hired to close. It is rarely a ChatGPT problem — most teams already have the seats. It is a configuration and connector problem: no shared system prompt, no MCP connections to the tools staff actually use, no prompt library for recurring tasks, and no guardrails around what the assistant can touch. Left unaddressed, this typically costs a few hours per person per week in manual copying, and it compounds as the team grows.

This guide covers what a proper setup involves, how the work is sequenced over a week, what it costs, and what to have ready before kickoff. Bharat AI Saathi's version of this is called Team AI Assistant (MCP-Connected); the offer page is at Team AI Assistant (MCP-Connected).

What Team AI Assistant (MCP-Connected) actually does

It is a branded Claude or ChatGPT Team assistant, configured for one specific team's daily work, connected via MCP (Model Context Protocol) to three of that team's existing tools — commonly HubSpot, Gmail, Google Sheets, Slack or Notion, capped at three by design so scope and monthly usage stay predictable.

In practice this means the assistant has a defined persona and system prompt built around your team's actual workflows (not a generic "helpful assistant" prompt), a library of roughly 30 tested prompts for the tasks staff run most — drafting client updates, summarizing a deal from CRM notes, turning a spreadsheet of leads into a prioritized call list — and read/write access to the three connected tools within role-based limits your team sets. It is not a chatbot bolted onto your website, and it is not a general-purpose automation platform; it is a working assistant your team opens the same way they'd open Slack, that already knows where your data lives.

For solo operators who don't need multi-tool connectors, there's a lighter $500 Custom GPT or Claude Project tier — a well-built persona, prompt set and knowledge base without the MCP connections.

How it works, step by step

1. Kickoff call. You walk through the top 5-10 tasks your team repeats weekly, which three tools matter most, and who should have access to what.
2. Access provisioning. You issue scoped API credentials or OAuth access for the three chosen tools — nothing broader than needed.
3. Persona and system prompt build. We draft the assistant's role, tone and boundaries based on how your team actually communicates with clients and each other.
4. MCP connector setup. The three tools are wired in, with role-based access so, for example, a junior staffer can read HubSpot but not send outbound email.
5. Prompt library build. Roughly 30 prompts are written and tested against your real data for the recurring tasks you flagged in step 1.
6. Acceptance test. We run 20 prompts live with your team present, adjusting wording and permissions on the spot.
7. Training and handover. Two recorded sessions plus a written SOP, then the assistant moves into daily use with the monthly care plan running in the background.

Trigger: Recruiter pastes a candidate's LinkedIn summary and asks "draft a client-ready blurb and log this candidate in HubSpot as 'Screened.'" Action: Assistant drafts a 4-sentence blurb in the agency's house tone, creates the HubSpot contact with the "Screened" stage set, and confirms back: "Blurb drafted below, contact logged. Want me to also add them to the Friday shortlist sheet?"

Before and after

What it costs and what is included

There's no lock-in: the setup fee buys a working system your team owns and operates; the monthly fee covers maintenance, not access. The payback logic is straightforward — if the assistant saves each connected staff member even an hour or two of manual copying and drafting per week, that typically covers the monthly care plan within the first few weeks, though actual time saved depends on how many people use it and how repetitive their tasks are.

Delivery timeline

Day 0: Kickoff

Discovery call, task prioritization, access requests sent.

Days 1-4: Build

Persona and system prompt drafted, MCP connectors configured and tested against your data, prompt library written.

Days 5-6: Acceptance testing

20-prompt live test with your team, adjustments to wording, permissions and connector behavior.

Day 7: Go-live

Training sessions recorded, SOP delivered, assistant moves into daily use.

What you need to have ready

• Active ChatGPT Team or Claude Team/Enterprise subscription with available seats
• Admin or owner access to the three tools you want connected (or someone who can grant scoped API/OAuth access quickly)
• A short list of the 5-10 tasks your team repeats most often
• Names and roles of who should have access, and what each role should and shouldn't be able to do
• Internal sign-off on connecting client data (CRM, email, spreadsheets) to an AI assistant — check with whoever owns data-handling decisions
• Availability for one kickoff call and one 30-45 minute acceptance test session

Doing it yourself vs done-for-you

DIY is a reasonable path if you already have someone comfortable with API credentials and prompt engineering and the time to iterate. Where it usually breaks down is maintenance — the initial build gets done, then nobody owns keeping it current as tools change or the team grows.

Compliance and risk notes

• Confirm your organization's policy on sending client or candidate data through a third-party AI platform before connecting any tool — this varies by industry and jurisdiction.
• A Data Processing Agreement (DPA) is available on request; check whether your compliance function requires one before granting access.
• Data residency depends on the underlying platform (OpenAI or Anthropic) and the connected tools — confirm this meets your requirements if you operate under GDPR, UK GDPR or similar regimes.
• Role-based access limits who can read or write via the assistant, but it does not replace your existing access controls on the underlying tools themselves.
• Platform terms of service for ChatGPT Team and Claude Team change periodically; the care plan includes monitoring for changes that affect your setup, but you remain responsible for your own platform account compliance.
• This guide is informational and not legal advice; consult your own counsel on data handling and jurisdiction-specific obligations.

Mistakes buyers make

• Connecting too many tools at once. Scope creep beyond three tools slows delivery and muddies guardrails — start with the three that carry the most repetitive work.
• Skipping the access-planning conversation. Deciding roles and permissions after the assistant is built means redoing guardrails — decide this at kickoff.
• Writing prompts around what the AI can do, not what the team needs. A prompt library should mirror actual recurring tasks, not generic capabilities — bring your real workflows to the kickoff call.
• No one owns the assistant after go-live. Without a designated internal point of contact, adoption stalls — assign someone before launch.
• Treating the setup fee as covering ongoing costs. Platform seats and API usage are billed separately and continue regardless of the care plan — budget for both.

FAQ

Do I need to already have ChatGPT Team or Claude Team set up?

Yes, an active subscription with available seats is a prerequisite, since the assistant is deployed inside that environment. If you're not yet subscribed, get that in place before booking a kickoff call.

Can we choose which three tools to connect?

Yes — the three MCP connectors are chosen based on where your team's daily work actually happens, commonly from HubSpot, Gmail, Google Sheets, Slack or Notion. The cap is fixed at three to keep scope and monthly usage predictable.

What happens if we want a fourth tool connected later?

That's a scope change handled under the monthly care plan or as a separate addition — it's not included in the base $1,500 setup.

Is this the same as building a custom chatbot for our website?

No. This is an internal productivity assistant for your team's own tools and workflows, not a customer-facing chatbot.

What does the monthly care plan actually cover?

Prompt and connector updates as your workflows change, a periodic usage review, and migration support when the underlying AI model updates — so the setup doesn't go stale.

How is this different from just writing our own prompts in ChatGPT?

Custom prompts alone don't connect to your tools or enforce role-based access. The MCP connectors, guardrails and tested prompt library are what turn a chat window into a working assistant tied to your actual data.

Next step

Email support@bharataisaathi.com or message WhatsApp with your team size, current ChatGPT/Claude plan, and the three tools you'd want connected.

See full scope and pricing at Team AI Assistant (MCP-Connected), or read how we work before booking a kickoff call.