MCP servers that let AI assistants use your product, with the permissions you set
An MCP server is a service that exposes your product's data and actions as tools that AI assistants can call through the Model Context Protocol, the open standard Anthropic introduced in November 2024 and that Claude, ChatGPT, Cursor, VS Code and other clients now support. Sysvoy builds MCP servers for SaaS products and internal systems: a small set of tools that models use correctly, OAuth sign-in scoped to each user, audit logs, and hosting as a remote server or inside your own network.
Where MCP server development projects stall
And what we do differently.
Every endpoint becomes a tool
Fifty vaguely named tools confuse the model and crowd its context. We design a small set of task-shaped tools and test that the model picks the right one.
Weak auth and leaky tools
A shared API key gives every user the same access, and broad tools return records nobody asked for. We use per-user OAuth tokens and return only the fields a task needs.
Never tried in a real client
A server that passes unit tests can still fail inside Claude or ChatGPT. We test in the clients your users have before anyone else sees it.
What we build
An MCP server for your SaaS
Your customers connect your product to Claude, ChatGPT or Cursor and work with their own data in it, under their own login.
Internal MCP servers
Your team's agents get access to databases, CRMs and document stores through tools you define, not through shared credentials.
REST API to MCP
We turn an existing API into a few task-shaped tools, rather than one tool per endpoint, and host it as a remote MCP server.
MCP gateways
One entry point in front of several MCP servers, with sign-in, rate limits, tool allow-lists and a log of every call.
Self-hosted MCP and models
MCP servers and open-weight models on your own servers, like the shopping assistant we built for a US B2B distributor, where customer data never leaves their environment.
How we deliver
From the first workshop to production. How we work
Tools designed for models
Few tools, clear names, plain descriptions and typed inputs. We check which tool the model picks for real requests and rewrite until it picks the right one.
Read-only first
The first release reads and searches. Write actions come one at a time, with a confirmation step wherever a mistake would cost money.
Scoped access with OAuth
Remote servers follow the MCP authorization spec, which is built on OAuth 2.1, so each user's token only reaches that user's data.
Tested in real clients, logged in production
We run the server in Claude, ChatGPT and Cursor before launch, and log every tool call with who made it and what it returned.
An MCP server compared with a REST API and a custom function-calling integration
| REST API | Custom function calling | MCP server | |
|---|---|---|---|
| Built for | Developers writing code | One AI app you build | Any MCP-compatible AI client |
| Works in Claude, ChatGPT and Cursor | Not directly | No, built per app | Yes, one server for all of them |
| User sign-in from AI clients | Your own scheme | Handled inside your app | OAuth flow defined by the spec |
| Best for | Connecting systems to each other | A single assistant inside your own product | Letting customers' and staff's AI tools use your product |
A good fit when
- Customers keep asking whether your product works with Claude or ChatGPT.
- Your team's agents need internal data, and handing them database passwords isn't an option.
- You have a REST API and want AI clients to use it without a separate plugin for each one.
What we build with
- TypeScript
- Python
- Official MCP SDKs
- Streamable HTTP and stdio
- OAuth 2.1
- PostgreSQL
- Typesense
- HubSpot & QuickBooks APIs
- Ollama and vLLM
- AWS and Google Cloud
Related work
AI commerce assistant
Self-hosted AI shopping assistant, Typesense product search and HubSpot integration for a US B2B distributor.
US distributormidpage.ai
AI research platform that layers analysis over case law for litigators and legal teams.
midpage.ai ↗AI security platform
AI-driven security platform and SOC tooling for a European cybersecurity company.
Client under NDA
Common questions
Not answered here? Email contact@sysvoy.com.
How much does it cost to build an MCP server?
Most MCP servers for an existing product cost $3,000–$8,000 and take 1–3 weeks. The price depends on how many tools you need, whether users sign in with OAuth, and whether we host it for you. A gateway across several systems, or self-hosted models alongside the server, costs more, and we quote it after a short scoping call.
How long does it take to build an MCP server?
One to three weeks for most products that already have an API. We start with tool design and a read-only version you can try in Claude or Cursor, then add sign-in, write actions, testing and deployment. If there's no API yet, we build that first and the timeline grows with it.
Which AI apps can connect to an MCP server?
Claude, ChatGPT, Cursor and VS Code all support MCP, and the list keeps growing. Claude and ChatGPT connect to remote servers over the internet with OAuth sign-in, while tools such as Cursor can also run a local server on the user's machine. We test against the clients your users actually have.
Do we need an MCP server if we already have a REST API?
Your API stays the foundation. The MCP server sits on top of it, describes a handful of tasks in a way models understand, handles sign-in for AI clients, and works in every MCP client at once. Without it, each AI integration needs its own custom code.
How do you keep an MCP server secure?
Remote servers use OAuth, so every request carries a token issued for that user and that server. Each tool reads only the data its task needs, the first release is read-only, every input is validated and every call is logged. We also test for prompt injection, because text returned by a tool can try to steer the model.
Can the MCP server run on our own infrastructure?
Yes. We deploy remote MCP servers to your cloud account or your own servers, and internal servers can stay inside your network. For a US distributor's shopping assistant, the models also run on their own servers with Ollama and vLLM, so customer data never leaves their environment.
Is MCP a stable standard to build on?
MCP was introduced by Anthropic in November 2024 and donated in December 2025 to the Agentic AI Foundation, part of the Linux Foundation. The spec is versioned by date and still changing, so we build on the official SDKs and keep servers current when a new version affects them.
Other services
- AI agent developmentProduction AI agents and multi-agent systems that take research, triage, document review and operations off your team.
- LLM & RAG developmentRetrieval pipelines grounded in your own data, with evaluation and citations so answers can be checked, not just trusted.
- AI for securityAI-driven security platforms and SOC tooling — and hardened LLM products that can't be talked into leaking data.
- SaaS product developmentFull-stack Next.js, Node and TypeScript products — web and mobile — from a validated MVP to a platform that scales.
- Integrations & cloud migrationSalesforce and HubSpot integrations, legacy systems connected to modern ones, and applications moved to the cloud.
- Dedicated development teamsSenior engineers from our Lahore studio who work inside your tools and rituals, with overlap hours agreed upfront.
Have something to build?
Tell us where things stand and what you need. A short message is enough to start.