“Where are we on the project?”
What shipped this week, what is waiting on their side, read from your tracker instead of an account manager's memory.
for professional services and agencies
Your agency, inside your clients' AI chats
Today a client emails to ask where their project stands, what the retainer covers and which invoice is open. With a connector, they ask their own assistant, and the answer comes from your CRM, your tracker and your books, inside the boundary you set. You keep the systems you already have. The first step is two questions and a free diagnosis.
the messages you answer by hand
“Quick one: where are we on the redesign?”
“Does the retainer cover the extra landing page?”
“Which invoice is still open on our side?”
Every one of them lands on an account manager who then opens three tabs to answer it. The client is not asking for a meeting, they are asking for a fact. A law firm and an accounting practice run the same list with different nouns: matters instead of projects, filings instead of deliverables.
your clients
What your clients do in the chat
They type in the assistant they already use. Your connector answers from your systems, with no login for the questions that need none.
“Can I get twenty minutes with whoever runs our account?”
A call with their own account manager against the team's real availability, booked in the chat rather than in a thread with a shared inbox.
“What does our retainer actually cover, and what is open?”
The money question answered without an awkward email: what the retainer includes, which invoices are open, when the next one lands.
“What are you waiting on from us?”
The access, file or approval that is holding the work, named out loud, and confirmed the moment it arrives.
“What does a discovery engagement involve?”
Scope, timeline and what falls outside it, answered from your own service descriptions and framed the way you frame them.
your team · signed in
Same connector, a second door
Behind a login, against the same client data, your account managers and partners see the side clients never do.
“Open my week.”
Active projects, deadlines this week, and which clients have not heard from you lately.
“What needs attention?”
Stale proposals, unbilled work and overdue invoices in one answer instead of three tabs.
“Context on this client before the call.”
History, open items, last invoices and who promised what, instead of scrolling three threads.
“Log that and nudge billing.”
A follow-up task, a note on the account, a nudge to billing, each change confirmed before it fires.
Your team edits in your tracker. Your client sees it in their own chat, with no step in between.
taskstate
Brand auditdone
Homepage copyin review drafting
Analytics setupblocked: access
Where are we on the redesign?
Homepage copy is in review with you. Analytics setup is waiting on your account access.
proof you can check
How do we know you can do this?
We sell yesmcp through our own connector. It is live, it works on any Claude plan, and it books a real consultation in a real calendar. Break it if you like: we read the logs.
Open our connector in Claude →We are a professional-services practice ourselves, which is why this page describes your week accurately. The connector you would buy is the one we run on our own business.
We counted the Claude connector directory by hand on 24 August 2026: 2105 connectors, 94 of them filed under legal, 26 under health. There is no category for a professional practice at all, because the directory is organised around software vendors rather than around firms like yours.
the risk question
What about the risks?
Risk is the right question. Most of this work is design rather than code, precisely because of it.
Will it see more than it should?
Scope is decided before anything is built. The client-facing tools only ever see what you would put in a status email. Internal notes, margins and team chatter are not hidden by an instruction, they sit outside the tools' reach entirely.
Can it change things on its own?
Reads and writes are separate tools, and every write is confirmed before it fires. On our own connector a booking is not committed until the confirmation goes out, so a sentence that merely sounds like a request changes nothing.
What if a client's file carries an instruction to the assistant?
This is real, and we published our own findings on it: a tool result is read by the model as text it should act on, which makes anything a client sends a possible instruction. The defence is structural. A tool that reads a client's document has no write capability at all, so an instruction hidden inside that document has nothing to call.
Who can see what happened?
Every tool call your connector serves is written to an audit log on infrastructure we run for you: which tool, which arguments, how long it took, what it returned. Today those logs are ours to read and yours to ask for.
Does this change our client contract?
Client data stays in your systems and is never used for training. Access is scoped per client, which is a design commitment we make per engagement rather than a switch we flip: your first scenario defines what each client may see, and the tools cannot exceed it. OAuth 2.1 is the pattern we run in production on our own login-gated service. If your engagement letter needs specific language, that conversation happens in the diagnosis, before anything is built.
your side of it
What does it take from your side?
- Wherever your client work lives, a CRM, a project tracker, an accounting tool or spreadsheets, needs an API or a small bridge to one. Answers that span several of them are the goal we design toward, and we say plainly which of your systems can carry that today and which cannot.
- A clear client-visible boundary, decided by you, so the connector cannot cross it.
- A decision on identity. Answering general questions needs nothing; showing a specific client their own projects and invoices needs a login step.
- A decision on what the assistant may promise. Sharing a status is safe. Committing to a deadline or a discount is not, and we design the tools so that boundary cannot be crossed by accident.
- About an hour of one person who knows your systems. First phase, that is the whole ask.
cost and time
What does it cost and how long does it take?
The diagnosis costs nothing
Two questions in a chat, and you get a map of which scenarios are worth building for your firm, in what order, and which of them your systems can carry today. You leave that step with a number, before you have spent an hour of your own time.
A fixed price, quoted for your firm
We quote per firm, because the number turns on the one thing a web page cannot guess: what your systems already expose. A tracker with a clean API and a spreadsheet that holds the real truth about scope are different jobs. Fixed means fixed: the scope is written down, and the price does not move with it.
A monthly line, quoted with the first
A connector is not a delivery. We host it, watch it, and keep it working as the platforms change. That number comes with the first one, so you see both before you decide anything.
If the first scenario turns out to be the wrong one, we build the next one from your map at no extra cost. The map exists precisely so that the choice is not a guess.
About two weeks for a first scenario against a system with an API. At the diagnosis we say plainly which of your systems can carry that today and which cannot.
Step zero is free and takes two questions: get your scenario map →
before you write
Questions professional-services firms ask
→What does my client need to use this?
Claude, on any plan, including the free one. They add your connector once, from a link you send, and from then on they just type in the chat.
→Does this work in ChatGPT too?
The same connector is built to the open standard both platforms speak. Claude accepts custom connectors by link today; ChatGPT's route is its directory, and we build so that one connector serves both as that opens.
→Do we have to replace our CRM?
No. The connector reads from what you already run. Replacing tools is the opposite of what this is for.
→What if a platform stops rendering the interactive cards?
The same scenario keeps working as text. We do not control platform features, so we design for one going away without the scenario breaking.
→Can we start with the team side instead of clients?
Yes. Same connector, a second door. Firms often find the internal side easier to approve, because no client sees it while you learn what it should do.
→Who owns the connector?
You do. It runs on infrastructure we manage, the code and the scenarios are yours, and we keep you by doing the work rather than by holding the keys.
→How long until the first scenario is live?
About two weeks, when the system holding the data has an API. If it does not, the bridge to it is the first thing we scope, and we tell you that at the diagnosis rather than halfway through.
→Why is there no price on this page?
Because the honest number depends on what your systems already expose, and we would rather tell you the real one than a safe one. The diagnosis is free, it takes two questions, and you leave it with the figure for your firm.
next step
Start with the map, not with a contract
Two questions about your practice, and you see which scenarios are worth building, in what order, and what your systems can carry today.
Get your scenario map →
Rather talk to a person? Message us on Telegram or write to [email protected]