New member
- Messages
- 8
Hi everyone,
I’m building Vectoralix, a managed platform for creating, testing, publishing and operating Model Context Protocol servers.
The idea came from a problem I kept running into while working with AI tools: the model is rarely the only limitation. The useful knowledge is usually somewhere else—in documentation, code repositories, internal procedures, APIs, files and the heads of people who have spent years doing the work.
You can paste some of that context into a chat, but that quickly stops working when the knowledge becomes large, changes regularly or needs to be shared across a team.
MCP gives AI clients a standard way to access external knowledge and tools. But building a real MCP server still means handling the protocol, hosting, authentication, tool execution, security, releases, logging and maintenance.
Vectoralix is intended to handle that infrastructure layer.
You can then attach tools that determine what an AI client can actually do with that information.
It can search content, list files and categories, explore related items and retrieve only the information needed for the current task.
A development team could use this for repository knowledge, architecture decisions, runbooks and internal documentation. A consultant could use it to make years of domain knowledge available through an AI assistant.
You configure the URL, parameters, headers, authentication, input mapping and response processing. The AI client can then invoke the API through the MCP server without needing to know how the underlying integration works.
This could be used to look up orders, create support tickets, check inventory, retrieve analytics or connect an internal business system.
The API layer includes protections such as SSRF checks, so it is not simply allowing an agent to request arbitrary URLs.
Vectoralix includes a Code Execute tool for running JavaScript inside a restricted QuickJS sandbox with execution time and memory limits.
That makes it possible to expose custom calculations and business rules without deploying another standalone service.
For example, a tool could calculate pricing, validate an order, transform API data, compare measurements or apply a company-specific formula before returning the result to the AI.
Vectoralix includes a Playground where you can inspect and test resources and tools before publishing the server. API calls and code execution can be tested and debugged directly from the dashboard.
Vectoralix uses immutable releases. You prepare a version, test it, activate it and keep a known-good fallback version available.
If something goes wrong, you can roll back without rebuilding the server or changing the MCP URL used by clients.
It can also be protected with bearer-token or OAuth access when it contains private company knowledge or internal actions.
The same platform can therefore be used for a public educational MCP server, a personal knowledge server or a private server shared by members of an organization.
This becomes especially important once an MCP server is being used by more than one person or included in an automated workflow.
But I think the more interesting applications may eventually come from industries that are not primarily software businesses.
A fishing company could combine years of knowledge from captains, vessel procedures, weather data and operational APIs.
A manufacturing company could expose machine instructions, troubleshooting procedures, material specifications and internal calculations.
A person could maintain a private, manageable knowledge base about their own work and life instead of repeatedly explaining the same background in new AI conversations.
A consultant could package specialist knowledge together with calculators and API tools, then make it available to selected clients.
The goal is not only to give AI more text. It is to provide structured knowledge and carefully limited actions.
As an example, I created a Healthy Food MCP server with tools for nutritional calculations, ingredient substitutions and metabolic calculations such as BMR and TDEE.
It is partly a working example and partly an experiment: I want to show that MCP is not limited to coding assistants and developer infrastructure.
More public Cookbook examples will be added as I explore different industries and use cases.
It is the layer between those clients and the knowledge, APIs and tools they need to use.
You configure the knowledge and capabilities once, publish an MCP endpoint and then connect it to compatible AI clients.
Vectoralix currently supports Streamable HTTP and integrations with clients such as Claude Desktop, Claude Code, Cursor and custom MCP applications.
You can explore the platform here:
https://vectoralix.com
Documentation:
https://vectoralix.com/docs
Public MCP directory:
https://vectoralix.com/mcp-servers
There is a free plan and no credit card is required to get started.
I would genuinely appreciate feedback from people already working with MCP:
What part of building or operating MCP servers currently takes the most effort for you? And what kind of knowledge or business workflow would you want to expose through an MCP server?
I’m building Vectoralix, a managed platform for creating, testing, publishing and operating Model Context Protocol servers.
The idea came from a problem I kept running into while working with AI tools: the model is rarely the only limitation. The useful knowledge is usually somewhere else—in documentation, code repositories, internal procedures, APIs, files and the heads of people who have spent years doing the work.
You can paste some of that context into a chat, but that quickly stops working when the knowledge becomes large, changes regularly or needs to be shared across a team.
MCP gives AI clients a standard way to access external knowledge and tools. But building a real MCP server still means handling the protocol, hosting, authentication, tool execution, security, releases, logging and maintenance.
Vectoralix is intended to handle that infrastructure layer.
How it works
You create an MCP server and bring in the knowledge it should expose. This can include:- Files and documents
- Content written directly inside Vectoralix
- GitHub repositories
- GitLab repositories
- Bitbucket repositories
You can then attach tools that determine what an AI client can actually do with that information.
File Search
The File Search tool allows an AI client to navigate the connected knowledge rather than loading the entire knowledge base into every prompt.It can search content, list files and categories, explore related items and retrieve only the information needed for the current task.
A development team could use this for repository knowledge, architecture decisions, runbooks and internal documentation. A consultant could use it to make years of domain knowledge available through an AI assistant.
API tools
Vectoralix can also expose external APIs as MCP tools.You configure the URL, parameters, headers, authentication, input mapping and response processing. The AI client can then invoke the API through the MCP server without needing to know how the underlying integration works.
This could be used to look up orders, create support tickets, check inventory, retrieve analytics or connect an internal business system.
The API layer includes protections such as SSRF checks, so it is not simply allowing an agent to request arbitrary URLs.
Sandboxed code tools
Sometimes retrieving information is not enough. The result needs to be calculated, transformed or validated.Vectoralix includes a Code Execute tool for running JavaScript inside a restricted QuickJS sandbox with execution time and memory limits.
That makes it possible to expose custom calculations and business rules without deploying another standalone service.
For example, a tool could calculate pricing, validate an order, transform API data, compare measurements or apply a company-specific formula before returning the result to the AI.
Built-in Playground
One frustrating part of building MCP servers is having to connect an external client before you know whether the tools work correctly.Vectoralix includes a Playground where you can inspect and test resources and tools before publishing the server. API calls and code execution can be tested and debugged directly from the dashboard.
Versions instead of live accidental changes
MCP servers can eventually become part of real workflows, so changing a tool or document should not unexpectedly break every connected client.Vectoralix uses immutable releases. You prepare a version, test it, activate it and keep a known-good fallback version available.
If something goes wrong, you can roll back without rebuilding the server or changing the MCP URL used by clients.
Public and private MCP servers
An MCP server can be public when its knowledge and tools are meant to be shared openly.It can also be protected with bearer-token or OAuth access when it contains private company knowledge or internal actions.
The same platform can therefore be used for a public educational MCP server, a personal knowledge server or a private server shared by members of an organization.
Logs and usage information
Vectoralix records MCP requests and usage metrics, making it easier to see whether clients are connecting, which operations they are performing and where something failed.This becomes especially important once an MCP server is being used by more than one person or included in an automated workflow.
What could people build with it?
The obvious use case is software development: connecting documentation, repositories, architecture decisions and developer tools to Claude, Cursor or another MCP client.But I think the more interesting applications may eventually come from industries that are not primarily software businesses.
A fishing company could combine years of knowledge from captains, vessel procedures, weather data and operational APIs.
A manufacturing company could expose machine instructions, troubleshooting procedures, material specifications and internal calculations.
A person could maintain a private, manageable knowledge base about their own work and life instead of repeatedly explaining the same background in new AI conversations.
A consultant could package specialist knowledge together with calculators and API tools, then make it available to selected clients.
The goal is not only to give AI more text. It is to provide structured knowledge and carefully limited actions.
Public MCP Directory and Cookbook
I have also started a public MCP directory where people can discover MCP servers published through Vectoralix or by third-party creators.As an example, I created a Healthy Food MCP server with tools for nutritional calculations, ingredient substitutions and metabolic calculations such as BMR and TDEE.
It is partly a working example and partly an experiment: I want to show that MCP is not limited to coding assistants and developer infrastructure.
More public Cookbook examples will be added as I explore different industries and use cases.
What Vectoralix is not
It is not another general AI chatbot and it does not try to replace Claude, ChatGPT, Cursor or other clients.It is the layer between those clients and the knowledge, APIs and tools they need to use.
You configure the knowledge and capabilities once, publish an MCP endpoint and then connect it to compatible AI clients.
Vectoralix currently supports Streamable HTTP and integrations with clients such as Claude Desktop, Claude Code, Cursor and custom MCP applications.
You can explore the platform here:
https://vectoralix.com
Documentation:
https://vectoralix.com/docs
Public MCP directory:
https://vectoralix.com/mcp-servers
There is a free plan and no credit card is required to get started.
I would genuinely appreciate feedback from people already working with MCP:
What part of building or operating MCP servers currently takes the most effort for you? And what kind of knowledge or business workflow would you want to expose through an MCP server?