The right choice when
- Developers spend hours answering "can you pull the numbers for…".
- You run your own product and want internal teams, or customers, to query it through an assistant.
- The questions repeat, so they can become tools.
Integration · PostgreSQL / MySQL + MCP
Your own database, answering in Claude. Without handing over a password.
An MCP server in front of your PostgreSQL or MySQL database, with named tools for the questions your team asks, so Claude and other AI assistants answer from your data without ever being given open query access.
- AI assistant
Claude, Cursor or your own agent asks a question.
- MCP server
Offers a short list of Your database tools and checks every call.
- Your database
The data stays here. Credentials stay on the server.
Named tools, not open SQL, by default.
01 / Is this the one
Is this the connection you need?
It suits a company with its own software and a database the team keeps querying.
Look elsewhere when
- You want the assistant to run any SQL it writes on production. We will not build that.
- The answers live in documents, not tables. That is a retrieval system.
- The database has no clear owner or schema documentation yet.
02 / What the work covers
From a list of questions to a server your team owns.
Most servers start with a handful of read-only database tools, tested in the assistant your team uses, then widened.
- 01
The questions first
We list what your team asks of database today and turn the common ones into tools with clear names and inputs.
- 02
Access and permissions
A database account or key with only the access the tools need, so each person sees what their role allows.
- 03
Server, log and limits
Input checks, sign-in, a log of every call and rate limits, hosted where you choose.
- 04
Tested, then handed over
Tested in your assistant, including the requests it should refuse. Code, deployment and documentation go to your team.
03 / Limits and risks
Where it can go wrong.
Open SQL is a risk
A model that writes its own queries can read more than it should or load the database. We use named tools on a read-only replica, and allow ad-hoc queries only where you choose, with limits.
Prompt injection is not solved
Text a model reads, such as a customer note or an email, can carry instructions. The server enforces its own rules and treats every call as untrusted.
An answer still needs a second look
The tools return exact figures and the assistant quotes them, but a model can still pick the wrong report or period. Anything you act on, check in the system itself.
04 / What we can show
An MCP server of our own, in public.
A public MCP server brings the same stories into Claude, Cursor and other AI assistants. Free beta. No account needed; a free key is optional.
What this does not show
A client project. FocusPulse is our own product, and we have no published case study of an MCP server built for a client.
05 / What to do next
Price it or talk it through.
The first step is a free scoping call about your database and the questions it answers today. A written estimate follows.
01
MCP development
02A controlled route into your data and tools.
Estimate the cost
03A price range for your build in four steps.
HubSpot and Claude via MCP
04Ask Claude about contacts, deals and tickets.
Tally and Claude via MCP
05Ask Claude about your Tally books.
All integrations
Every connection we build.
A little clarity
Before connecting a database to an AI assistant.
Can we see it working?
Yes. This connection is built and tested by our team, and we can demo it for you on a free call before you decide anything.
What is MCP, in plain words?
The Model Context Protocol is an open standard for connecting AI assistants to data and tools. A server built once can be used by any assistant that supports it, such as Claude or Cursor, so you are not tied to one AI vendor.
Does our data go to the AI company?
Only what a tool returns for a question, and only to the assistant you connect. Nothing is copied in bulk. Which assistant and which plan you use decides how that provider treats the conversation, so we choose that with you.
Which databases can you connect?
PostgreSQL and MySQL most often, and SQL Server, MongoDB and others in the same way. The database stays where it is; the server needs network access to a read-only copy or user.
What does it cost?
It depends on how many tools the first version needs and how the database is hosted. First releases at NSDBytes start from $5,000. The project cost estimator gives a range, and a fixed quote follows a free scoping call. Charges from the services themselves, such as WhatsApp messages, API plans and hosting, are billed by those providers and are separate.
Who owns the connection once it is built?
You do. The code, the server it runs on and the accounts with each provider are yours once the final invoice is paid. We can run and maintain it afterwards if you want, but you are never locked in.
Start wherever you are
Bring the questions your developers keep answering.
We'll show you which ones can become safe tools for an assistant.
Want certainty before you commit? Start with a scoping project, $500