Let ChatGPT & Claude work inside your product.
MCP is the open standard AI assistants use to read data and take actions in other software. We build the server that turns your product into one of the tools they can use, with your rules on what they may see and do.
Three parts, one protocol.
The assistant never touches your database. It talks to your MCP server, and the server decides what it may read and do.
Where the request starts.
A customer or a teammate asks a question or gives an instruction in the assistant they already use.
The part we build.
It offers the assistant a fixed set of tools and resources, checks who is asking, runs the request against your product and logs it.
Where the answer lives.
Your database, API, store or CRM. It stays where it is and keeps its own rules.
Actions the assistant may take
Search orders, create a task, update a record, start a refund. Each one is defined, documented and scoped by us.
Data the assistant may read
Product catalogues, documentation, account details, reports. Served live from your product, not copied into the model.
Who may call what
Sign-in through OAuth, a scope per tool, confirmation before writes, and a log of every call. Most clients start read-only.
The server, the connectors and the APIs behind them.
One team for the whole chain: the MCP server, the integrations around it, and any API your product is still missing.
Your product as a tool.
A server that exposes the right actions and data from your product, and nothing else.
- Tools for the actions that matter: search, create, update, approve
- Resources for the data assistants may read
- Remote servers for ChatGPT, Claude and Copilot, local servers for developer tools
- Built on the current MCP specification, kept up to date as it evolves
Plugged into what you already use.
Your product inside the assistants your team and your customers already open every day.
- ChatGPT connectors and Claude integrations for your product
- Shopify, Stripe, HubSpot, Notion, Slack and the rest of your stack
- Webhooks that let events in your product trigger assistant actions
- Existing MCP servers wired into your own app or agent
The foundation, if it is missing.
If your product has no API yet, we build the one the server needs, and you keep it.
- REST or GraphQL APIs on Node.js or Firebase
- OAuth sign-in, scopes and rate limits
- Audit logs for every assistant action
- Documentation and tests, so it stays maintainable
What people ask, once it is connected.
The useful MCP servers are small. A handful of tools that answer the questions your customers and your team ask every day.
Orders, returns and stock
Customers get order status, delivery changes and return steps inside their own assistant, read live from your store.
Records without the UI
Your users create and update records from ChatGPT or Claude instead of opening your app, with their own permissions.
Context before the reply
Support agents get the customer's history and a drafted answer from your own data, not from a generic model.
Live numbers, approved actions
Teams ask for figures and run approved actions without waiting for a dashboard to be built.
Our own MCP server, in Claude today.
We built the MCP server for SelfManager.ai. Users connect Claude and work with their real tables in plain words.
- Permissions chosen per connection. Read only, make changes, or also write in tables other people can see.
- Nothing deleted behind your back. It can’t delete tables, tasks, notes, comments or the journal, share tables or set deadlines.
- Every action visible. Each change is marked “(via Claude)” in the table’s log, and an activity page lists the last 30 days.
- Costs kept predictable. Reading a day or ticking a task is free; heavier requests use one AI credit each.
- Tested with Claude. Claude Code, Cursor, VS Code and LM Studio are next, and ChatGPT is coming. Currently in private beta.
From selfmanager.ai/mcp ↗
Scope small, ship, then extend.
A first version with a few well-chosen tools beats a large one nobody trusts. Usage shows what to add next.
Scope
Which actions and data to expose, to whom, and under which permissions. We pick the few tools that carry real value first.
Build
The server, the sign-in, the tools and resources, and any API endpoint your product is missing.
Test
With the real assistants and real prompts, including the requests that must be refused.
Launch & iterate
Deployed on your cloud with logs and versioning, then extended as usage shows what people actually ask.
Tell us what your product does and who should use it from an assistant. You get a scoping note and a fixed quote.
Get a quote →Built by people who run software, not just write it.
An MCP server is a product surface, with the same care for security, uptime and maintainability as the rest of your app.
The founder builds it
Marian Sorca scopes, leads and codes every project. Nothing is handed off, so the person who scoped your server is the one who ships it.
We run our own SaaS
We built the MCP server for SelfManager.ai, our own SaaS, built and run since 2016. We know what it takes to keep an API safe and online for real users.
One team, the whole chain
The MCP server, the API behind it, and the web app or Shopify store that goes with it, from a single team.
Before you ask.
The questions we get in the first call. Anything else: the contact form is the fastest way to reach us.
What is an MCP server?+
MCP, the Model Context Protocol, is an open standard for connecting AI assistants to software. An MCP server is the small service that sits in front of your product and offers the assistant a defined set of tools (actions) and resources (data). The assistant calls them, and the server runs them with the user's permissions.
Which assistants can use it?+
ChatGPT, Claude, Microsoft Copilot, Cursor and most agent frameworks support MCP, so one server serves all of them. Your own in-app assistant can use it as well.
Is it safe to let an assistant touch our data?+
The assistant only gets the tools you expose. Sign-in uses OAuth, each tool has its own scope, writes can require a confirmation, and every call is logged. Most clients start read-only and add actions once they trust the results.
Do we need an API first?+
No. If your product has an API, we build the server on top of it. If it does not, we build the endpoints the server needs, and the API stays yours afterwards.
Who hosts it, and who owns the code?+
It runs on your cloud, usually Firebase or Cloud Run, Vercel or AWS, under your own account. You own the code and the logs.
How long does a first version take, and what does it cost?+
Weeks, not months, for a focused first version with a handful of tools. We scope it in the first call and send a fixed quote, so you know the price before anything is built.
Have a project in mind?
Tell us what you’re building and where it’s stuck. Within 24 hours you get useful questions and a proposed next step, or an honest answer if we’re not the right fit.
- 01Tell us about it.A few lines on the goal, the deadline and what exists today.
- 02A reply within 24 hours.Useful questions and a proposed next step, in writing.
- 03Scope and price agreed before work starts.Projects from $2,000. Ongoing fractional CTO work is billed monthly.