Service
Custom MCP servers for your own systems
An MCP server lets AI assistants such as Claude and ChatGPT look things up and take actions in your own systems, with the permissions you choose. We design, build, secure and host them.
What we build
- Read-only tools first: search your products, orders, tickets or documents from an assistant
- Write actions second, with confirmation steps for anything that changes data or contacts customers
- Remote servers over Streamable HTTP for teams, or local servers for a single machine
- Authentication with OAuth 2.1 or API keys, and validation of the Origin header
- Audit logs of every tool call
- Testing end to end with the assistants and IDEs you use
Designing tools an assistant uses well
Most of the work is deciding which tools to expose and how narrow to make them. A few clearly named tools with tight inputs give better and safer results than one tool that can do anything. We start from the questions your team actually asks, and add actions only once the read-only tools are trusted.
Hosting and handover
We deploy remote servers on your infrastructure or ours, behind a proxy configured for streaming, with health checks and resource limits. You get the source code, the deployment configuration and documentation of every tool. Model and assistant subscriptions, and changes to your own internal APIs, are not included.
Why us
We build and run our own MCP servers in production on Kubernetes, and we documented what it takes in our engineering note on hosting remote MCP servers. Projects are quoted after a scoping conversation: in ringgit for Malaysian clients and in US dollars for clients elsewhere.
Frequently asked questions
Which AI assistants can use an MCP server?
A growing list of assistants and developer tools support MCP, including Claude and ChatGPT. We test against the specific clients your team uses before handover.
Is it safe to connect an AI assistant to our systems?
It can be, with the right design: least-privilege tools, authentication, audit logs, and confirmation steps for any action that changes data. We start read-only and add write actions deliberately.
Local or remote MCP server?
A local server suits one person on one machine. A remote server suits a team or hosted assistants, and keeps credentials for internal systems on the server instead of on every laptop.