Fill in a guided wizard and get a ready-to-run MCP server. Claude and other AI assistants can then search your products, check stock and prepare quotes, with controlled access and every action logged.
MCP (Model Context Protocol) is the open standard Claude, ChatGPT, Microsoft and Google use to connect AI to real systems.
company: name: Blue Harbor Supply defaultCurrency: USD server: name: blue-harbor-sales transports: [stdio, http] api: baseUrl: https://api.blueharbor.example/v2 auth: type: api_key valueEnv: BLUE_API_KEY # never typed in tools: [search_products, check_stock, get_pricing, create_quote] access: apiKeys: - id: harbor-sales-team scopes: [catalog:read, stock:read] requestsPerMinute: 60 audit: { enabled: true }
Loading the video…
A sample company, Blue Harbor Supply, fills in the 8 steps, generates its server, builds it and tests it. Tap a chapter to jump there.
The wizard validates every answer on the spot and shows the manifest it is building. If something is missing, it tells you what and on which step.
Name, description, instructions for the AI, and how apps connect.
Point it at your existing API. The key is read from an environment variable, never typed in.
Product search, stock, pricing, quotes and order status.
Map your API's field names to standard ones so the AI sees clean data.
Add actions unique to your business.
Share documents like your return policy, plus reusable prompts.
Keys per team, what each may do, per-minute limits and the audit log.
One click on Generate server and you get the code plus the next steps.
npm install npm run buildEvery generated server and the web dashboard ship with security built in. Nothing to bolt on later.
The dashboard uses your Microsoft account (Entra ID, formerly Azure AD). Only invited people get in.
Each client gets its own workspace. Client B cannot read, change or download anything of client A's.
Your API keys live in environment variables. Access keys are stored only as a SHA-256 hash.
Decide what each team may do: for example, read catalog and stock but not place orders.
The audit log records every tool call, and you can redact sensitive fields.
Manifests are encrypted at rest (AES-GCM). Usage is rate limited per client and per sign-in.
Security tests pass on the dashboard, plus 71/71 generator tests. Includes: forged sessions refused, uninvited users blocked, client isolation, and verified token signatures.
Monthly prices in US dollars.
For a small business with one system.
For teams with several departments or brands.
For agencies connecting their own clients.
One-time. We set up your whole connector.
Larger companies that need it in their own cloud: custom enterprise license.
Today's tools are built for developers. This one is built for business owners, in English and Spanish.
of enterprise apps will include task-specific AI agents by the end of 2026, up from less than 5% in 2025.
Gartner, August 2025MCP has been part of the Linux Foundation's Agentic AI Foundation since December 2025, backed by Anthropic, OpenAI, Microsoft, Google and AWS.
MCP blogWe're taking a small number of pilot companies at a discount. We'll walk you from step one until the AI answers with your data.