Your users already ask an AI
Your customers and their teams write, plan and analyse in Claude and ChatGPT every day. Today they leave it to find anything in your product.
MCP server development · Connector Starter · $2,500
Your product, inside Claude and ChatGPT.
Your users already ask an AI assistant about their work. We connect the product you have to Claude and ChatGPT, so they can ask about their own data and get answers from your product, under their own login and your rules. No new app, no model to host.
Your login
Each user signs in with their own account
Read-only first
Three tools that change nothing
$2,500
One time, with a two-week warranty
AI working through existing APIs and logins at
Your AI assistant
Claude or ChatGPT
Which of my customers have overdue invoices?
Illustrative example on a fictional product. Not client work.
Why connect
Your users ask an AI first. Your product should answer.
MCP, the Model Context Protocol, is the open standard AI assistants use to reach other software. An MCP server puts a few clear tools in front of your API, and the assistant calls them for the user. Read how tool calling works.
Your product answers there
They ask in plain words. The assistant calls your product, as that user, and the answer comes back where they are already working.
Only what each user may see
Each person signs in with your own login. Your product still checks every request, so a rep sees their accounts and a manager sees the team.
No new app to build
No chat screen, no model to host and no model bills of your own. Your users run it on their own Claude or ChatGPT plan.
Build a copilot or connect
The smaller first step. Nothing thrown away.
A copilot inside your product is sometimes the right call. When your users already work in an AI assistant, connecting comes first, and the same tools serve your own copilot later.
Build a copilot in your app
A new product to design and run.
- New screensA chat panel, streaming, approvals, all designed and supported
- Model costsYou pay for every model call your users make
- What gets builtAn agent, prompts, model hosting and a user interface
- Where users workOnly inside your product
- First versionA project of several months
- What it leads toYour own AI feature
Months of work before anyone asks a question.
Connect your product with Datics
Your product, where users already ask.
- New screensNone. Claude or ChatGPT is the screen
- Model costsUsers run it on their own Claude or ChatGPT plan
- What gets builtA few clear tools over the API you already have
- Where users workIn the assistant, next to the rest of their work
- First versionThree read-only tools, targeted for one sprint
- What it leads toThe same tools can power your own copilot later
- Your login
- A few clear tools
- Runs where you run
Try it
Ask it as different people.
Pick who is asking and what they ask. The owner sees the whole company, a dispatcher sees one region, a customer sees only their own account, and the connector itself changes nothing.
1 · Ask as
2 · Pick a question
The same connector and the same product. Each person signs in with their own Fieldline login, and Fieldline decides what each of them may see.
Your AI assistant
Fieldline connector
Signed in to Fieldline as Rosa, owner · the whole company
Which invoices are overdue?
Used Fieldline · list_overdue_invoices
Answered as this user
Four invoices are overdue across the company, $32,650 in total. Harbor Foods is the oldest.
- Harbor Foods$18,400 · 41 days
- Lakeside Properties$6,250 · 19 days
- Pine Street Clinic$4,900 · 12 days
- Orchard Schools$3,100 · 6 days
Read-only · logged with the user, the tool and the result
Illustrative example on a fictional product. Not client work.
How it runs
Three tools your users need most. Then live.
Four steps, from a free call to a connector your customers turn on. Every step ends with something you keep, and the price is fixed before the build starts.
30 minutes
No chargeProduct Review
Which questions your users ask most, and a first look at your API, your sign-in and your roles.
You keep
A straight answer on fit
Readiness check
Fixed priceTools and quote
We pick the first three tools with you, write what each one may return, and confirm the price or say what is extra.
You keep
A tool list you approve
One sprint, targeted
$2,500Build and test
The connector, sign-in, limits and logs, in your repository. Then real questions and the access tests that must fail safely.
You keep
A working test connector
Handover
Two-week warrantyLive and yours
Deployed where you choose, documented, and your team trained. A guide for your customers on turning it on.
You keep
Runbook, tests and a customer guide
Sample tool contract
list_overdue_invoices
- Does
- Overdue invoices for the customers this user can see
- Asks for
- A date range, 90 days at most
- Calls
- One known API request, as the signed-in user
- Returns
- Customer, amount and days overdue; no contact details
- Limit
- 25 rows
- Who may use it
- Owners, dispatchers and customers
- Tested
- 6 allowed, 4 denied, 2 errors
Your team approves each tool before it is built. One tool, one API call, only the fields the task needs.
Illustrative example on a fictional product. Not client work.
Security
Six tests. No handover until each one passes.
Your product stays the judge of who sees what. The connector can only narrow access, never widen it, and these cases are tested in your staging system before anyone uses it.
One user asks for another user's record
Your product denies it, and nothing of that record appears in the answer or the logs.
One customer types another customer's id
Denied, whatever id arrives. The customer is read from the sign-in, never from the chat.
A user calls a tool they may not use
Denied, even if they know its name. Nothing is allowed unless a rule allows it.
Access is removed
The next request is refused. You can switch off one user, one tool or one customer without a release.
Your API is down or slow
A clear error comes back. No made-up answer, and no request is sent twice.
A response holds a sensitive field
Only approved fields leave your product, with rows and size capped. Every call is logged.
The offer · Connector Starter
Your product answering inside Claude and ChatGPT, from $2,500.
Up to three read-only tools your users need most, each user signed in with their own account, tested in your staging before anyone uses it.
$2,500
Connector Starter · Fixed scope
What you keep
- A working connector with up to three read-only tools
- Each user signed in with their existing account and role
- Security tests that block a release if they fail
- A runbook and a handover to your team
A two-week warranty after release. The connector runs in your own cloud by default; managed hosting is optional.
It fits if you have
- A live product whose users already work in Claude or ChatGPT
- Data your users look up every day
It is not for
- Actions that change data: that is Connector Plus, from $5,000
- Products without a sign-in or an API
Want the AI inside your own screens instead? The Build Sprint builds it there, and it can reuse the same tools.
When you want more
Start with three read-only tools. Add more when you want.
One-time prices, no required subscription, and the code and setup are yours to run. Larger setups across several systems are quoted per project.
Connector Starter
$2,500one time
Your product in Claude and ChatGPT, read-only, for the questions your users ask most.
- Up to three read-only tools
- Your existing sign-in, per user
- Access tests, logs and limits
- One deployment, runbook and handover
- Two-week warranty
Connector Plus
From $5,000one time
From answers to action: users change records in your product, and each change waits for their approval.
- Actions that create or update records
- An approval before every change
- Interactive cards and forms in the chat
- More tools for more of your users' work
Managed hosting
From $299a month
Optional. We run and monitor the connector for you. Self-hosted in your own cloud is the default, with no monthly fee from us.
- Monitoring in business hours
- Updates and patches
- Usage limits stated up front
- Infrastructure passed through at cost
Is your product ready?
- A documented, working API for the tasks users ask about
- A staging system with at least two test users
- A sign-in that identifies each user and each customer
- A backend that checks permissions on every record
- Data you are happy for an AI assistant to show your users
Quoted on its own
A missing API, a new single sign-on build, several identity providers, or restricted and regulated data. We tell you in the readiness check, before you commit.
Book a connector callWe use it ourselves
A connector we run in our own operations.
Our team asks Claude and ChatGPT about projects, people and invoices through a connector we built. Each person signs in with their own account and sees only what their role allows. The same pattern runs in our client work.
Our own operations system
Live in Claude and ChatGPT. Each person signs in with their own account, keeps their own role, and can make only the updates that role allows.
In daily useCleanHaus
An agent that works through CleanHaus's own APIs, with a tool list per role and each user's own login.
Case studyOCCI
Live job cost data brought into Microsoft Copilot through a custom connector, with every answer checked against the source.
Case studyQuestions
MCP servers and Claude connectors, answered.
Planning a wider AI roadmap? See AI consulting.
What is an MCP server?
MCP, the Model Context Protocol, is the open standard Claude, ChatGPT and other AI assistants use to connect to outside tools. An MCP server sits in front of your API and offers the assistant a few clear tools, such as listing a user's overdue invoices. We build that server for your product, with your sign-in and your rules.
How much does an MCP connector cost?
The Connector Starter is $2,500, one time: up to three read-only tools, your existing sign-in, access tests, one deployment, handover and a two-week warranty. Connector Plus, for actions users approve and in-chat apps, starts at $5,000. Managed hosting is optional, from $299 a month. The price is confirmed after a short readiness check.
Does it work with both Claude and ChatGPT?
Yes. The connector follows the open standard both assistants use, and we test it in the one your users rely on, or both. Users need a plan that allows custom connectors, and on team plans an admin turns it on once for the company.
What data goes to the AI?
Only what a tool returns. That result enters the user's own chat, under their Claude or ChatGPT account and that provider's terms. So each tool returns only the fields the task needs, sensitive fields can be left out entirely, and you approve what is shared. We also give you a plain page for your customers on what the connector shares.
Can it change data in our product?
Not in the Starter, which only reads. Connector Plus adds actions that create or update records, and each one asks the user to approve it before your product makes the change. Your API checks their rights again on every action.
Who owns the connector and where does it run?
Your product, your API and your data stay yours. The connector runs in your cloud by default and your team owns its operation, or we run it for you from $299 a month. The deliverables and your rights to them are set in the statement of work.
What if our product has no API, or a sign-in that cannot be used?
Then we tell you in the readiness check, before you commit. A missing API, a new single sign-on build or several identity providers are quoted separately, so the Starter price stays fixed.
Why not build a copilot inside our product instead?
Sometimes that is the right call, and we build those too. But if your users already work in an AI assistant, connecting is the smaller, faster first step, and nothing is lost: the same tools can power your own copilot later.
Will our product be listed in Claude's or ChatGPT's directory?
Your customers can add the connector as a custom connector with no listing at all. If you want a directory listing, we prepare the submission, but Anthropic and OpenAI decide what they list.
Have you built one before?
Yes. We run one in our own operations: our team asks Claude and ChatGPT about projects, people and invoices through a connector we built, each person signing in with their own account and seeing only what their role allows. And we have built AI that works through clients' existing APIs under each user's own login.
Want your product answering inside Claude and ChatGPT?
Show us the questions your users ask most. In 30 minutes you will know whether your API and sign-in are ready, and what the first three tools would be.