MCP practice · scan free · you own the code

Your AI agents can't see your data. Find out how close they are.

The free scan reads your public DNS and website and maps every tool it confirms against the MCP ecosystem — what's agent-reachable today, what isn't, on screen in about a minute. Where there's a gap, that's the part I build: Model Context Protocol servers, scoped to your systems, after close to a decade building exactly this kind of integration.

No email required, nothing stored beyond the result — or skip straight to a project.

Your AI agentClaude · ChatGPT · in-housetools it can callMCP serverreaches only what you authoriseCRMWarehouseInternal APIyour systems, unchanged
3MCP servers I run in production
100%of the code is yours
$0platform fees after handover

Who builds it

I'm Luke Wade — a sales engineer who has spent close to a decade making integrations run against real customer systems, and who ships MCP servers as a habit, not a pivot. Three I run in production:

  • Memore — a private memory bank served to every AI tool over a remote MCP server, with OAuth 2.1 and its own vault app.
  • An internal CRM server — full read/write over every object in Insightly, driven from plain English, with in-conversation credential elicitation so keys never land in a chat log. Shipped to colleagues as a one-click desktop extension.
  • A self-hosted research agent — a multi-round planner/searcher/synthesiser with a local RAG library, running entirely on my own hardware.

That's the whole pitch: the person auditing your agent boundaries has built the thing he's auditing, more than once, including the parts that go wrong.

One practice, four rungs

Each rung stands on the one before it — and each is a fine place to start. Every price is on this page.

01 · Scan Free

Your public DNS and website, mapped against the MCP ecosystem: which tools you verifiably run, and how many an agent can already reach. No email, no account.

Run the scan

02 · Audit $3,500

The scan reads public records; the audit reads your actual stack. Auth boundaries, what agents should and shouldn't reach, which gaps matter, and what each costs to close. Fixed fee, report is yours either way.

03 · Build from $7,500

The gaps, closed: your CRM, warehouse, or internal API exposed to an agent as tools it can call — with you deciding exactly what it's allowed to reach. Built by someone with close to a decade of making integrations run against real customer systems.

See the process

04 · Assure $2,500

MCP servers are a new attack surface — tool poisoning is a named class now. I review deployments you already run, yours or a vendor's: what the agent can actually reach, where the auth is soft, what to fix first.

What a review covers

Fixed fees, agreed in writing before work starts. No retainers, no platform fees, no per-seat licensing. If scoping shows the work is smaller than the price, the price comes down; if it’s bigger, you get the number before anyone commits.

You don't have to cut anything first

Everything above works on the stack you already have. That said, wiring up four tools that do one job costs more to build and more to keep running — if you suspect you're paying twice for something, the free spend diagnosis next door is worth the detour.

Run the free audit

Three separate tools, one operator.

Each is bought, priced and run on its own, and each is worth doing by itself — none of them is a prerequisite for another, and there's no order you have to follow. Doing them together does compound: a bill you've already cut is a cheaper, cleaner thing to connect an AI agent to. That's an option, not a gate.

Got a system your agents should be able to reach?

Start with the scan — it's free and it makes the first conversation concrete. Or skip straight to telling me what's in your stack; if it's a fit, I'll come back with scope and a fee. If it isn't, I'll tell you that too.