Get Started

Your First MCP Server

A realistic first configuration: multiple content records with categories and a relation, a custom Code Execute tool you author and test, an explicit version cut with a release note, and a connection from a real AI client over the Bearer-authenticated private path.

What you will build

A private MCP server with three content records, one category, one relation, File Search enabled, one tested Code Execute tool, and version 1 active. By the end of this page, an AI client running on your machine will be able to call the server with a Bearer token and exercise both File Search and your custom tool.

You will need: a Vectoralix account, ten minutes of content (any text), and an MCP-aware AI client. Claude Desktop is the default example here; if you do not have it yet, the Playground inside the dashboard is a fallback.

Step 1 — Create the server (private)

Create a new MCP server as you did in Quick Start, but set Visibility to Private this time. Private servers reject anonymous traffic and require a bearer token on every request.

Plan tier matters here: cheaper plans cap the number of private servers you can run (typically one). The dashboard surfaces this limit before you save.

Step 2 — Add content three ways

On the Contents tab, create three records using each of the three sources Vectoralix supports:

  1. A Typed content record — paste a paragraph straight into the editor.
  2. A File upload — drop in a small Markdown file.
  3. Another Typed content record — useful for showing how relations connect two records.

All three are stored as database records keyed by a unique slug. There is no metadata layer and no object storage.

Step 3 — Organise with a category and a group

  1. On the Categories tab, create a category called "Guides".
  2. Back on Contents, assign all three records to the Guides category.
  3. On the Groups tab, create a group, then add two of the three records to it.

Categories label content; groups bundle records logically. File Search uses both to give AI clients narrow, contextual slices of your knowledge base instead of a flat list.

Step 4 — Define a relation

On the Relations tab, click Create and add a "depends-on" relation between two of your records (source and target). Save.

Heads up: Relations are immutable once created. To change one, delete the relation and create a new one.

Step 5 — Enable and test one Code Execute tool

  1. Open MCP Tools from the sidebar and create a new Code Execute tool.
  2. On the Configure page, paste a small JavaScript body — for example, a function that takes a string parameter and returns it uppercased.
  3. Add one parameter row through the Parameters relation manager so the tool exposes a typed input to AI clients.
  4. Click Test Code in the page header. Enter a sample value, press Run, and watch the result panel update.
  5. Once it works, attach the tool to your server from the server's Details tab.

Code runs inside a QuickJS sandbox with a configurable timeout and memory limit. The Test Code modal stays open between runs, so you can iterate on inputs without reopening the dialog.

Step 6 — Cut version 1 with a release note

  1. On the server's Versions tab, click Create Version.
  2. Add a short release note describing the initial state ("Initial guides, depends-on relation, Code Execute uppercase tool").

Version 1 is auto-active and auto-default. The version snapshot freezes the content, categories, groups, and relations as they are right now. Tools are not versioned — configuration changes to the Code Execute tool you just built take effect immediately on every server it is attached to.

Step 7 — Issue a Bearer token and connect from an MCP client

Generate a Personal Access Token from your account area, then point an MCP client at the server's endpoint URL with that token in the Authorization header. The endpoint shape is always: 

POST https://vectoralix.com/mcp/<serverUid>
Authorization: Bearer <your-token>

In Claude Desktop, open the MCP servers settings and add a new entry with the URL and Bearer token. The exact field labels vary by client, but the two values are always the same.

No client yet?: Use the Playground inside the dashboard for now. Quick-connect handles the token transparently and gives you the same end-to-end view a real client would see.

Step 8 — Verify from the Playground

  1. Open MCP Playground and Quick-connect to the server.
  2. Run full_text_search to confirm File Search returns the right record.
  3. Run your Code Execute tool with a sample input and check the result.
  4. Open the dashboard's metrics view; the requests you just made should show up in the per-server counters.

What to do next

  • Connect a Git repository so content stays in sync with your codebase (Source Integrations group, coming soon).
  • Add an API URL tool to call an upstream HTTP API from inside an AI conversation (Built-in Tools group, coming soon).
  • Cut a second version, change something, and switch the active version to see the rollback story end-to-end.
  • Read Core Concepts — a one-page glossary of every term you saw above.