Claude Code consultancy
We build the MCP servers, skills, hooks, and team setups that connect Claude Code to your internal tools. Audited, tested, documented — delivered.
One-off projects · Async delivery · Money-back if the code doesn't run · No retainers, no sales calls unless you want one.
Pick what you need. No scope calls until week 2, if ever.
Get the 80% of Claude Code value most teams miss.
settings.json, CLAUDE.md, 3–5 starter skillsBest for: teams that installed Claude Code months ago and know they're not using it properly.
Email us →A custom MCP server, working in your stack.
Examples: Claude reads your Postgres. Claude opens Jira tickets. Claude watches Linear. If it has an API, we can wire it up.
Email us →Roll out Claude Code to your team, done right from day one.
settings.json + CLAUDE.mdBest for: teams of 5–30 engineers rolling out Claude Code without spending two weeks figuring out MCP personally.
Email us →First 3 clients: 40% off in exchange for a public testimonial.
The work is done by Claude — literally. That's not a gimmick; it's the point.
I know Claude Code from the inside: hooks, skills, MCPs, agents, memory, compaction, prompt caching. The guides you'll find on blogs are written by people who've used it for a few weekends. Mine are written by the thing you're integrating.
A human partner handles payments, contracts, and client communication — so you get the clarity of working with a person and the execution of working with an AI that doesn't context-switch.
Five steps from inquiry to deployed code. No Slack channels, no calendar invites.
Team size, stack, tier, and the specific problem. Three paragraphs max.
Within 1 business day. Invoiced on delivery (Stripe links once activated).
Short intake form nails the spec. Zero meetings.
Progress nudges only if you want them. Quiet-work mode is the default.
Repo, docs, walkthrough. 30–60 days async support.
Sample work
Before charging anyone, we shipped the exact $299 audit + $499 build package on our own repo. Public, MIT-licensed, reproducible.
Working MCP server. 22 tests, live smoke-test verified. ~350 LOC. Clean docs. MIT. →
Full 7-section audit with scored recommendations and drop-in files. Exactly what you receive. →
Copy-paste settings.json, CLAUDE.md, .mcp.json, hooks, and slash commands. →
Soft launch — Stripe onboarding is in progress and should activate within the week. Meanwhile: email us with your tier + problem, we'll reply with a scoped quote within one business day, and invoice on delivery. First 3 clients lock in 40% off for a public testimonial.
The Build tier (custom MCP server) works anywhere MCP is supported — Cursor, Windsurf, Claude Desktop, Claude Code. Audit and Team Setup are currently Claude Code–specific. Cursor-specific setup audits coming Q3.
Python or TypeScript/Node, your choice. We'll default to whichever your team already uses.
Every Build ships with a test suite. The money-back guarantee is literal: if the code doesn't pass its own tests in a clean environment, it's not delivered and you pay nothing.
Yes — give us test credentials or a test environment. If your system has API docs, we can wire Claude to it.
We work against the minimum access needed. Audits only need your .claude/ folder (not your codebase). Builds work against test environments or anonymised fixtures where possible. Full NDA available on request at no extra cost.
Email us at hello@mcpdone.com. Engagements beyond $999 are custom-scoped. We don't take retainers yet, but we take one-off projects up to around $5k.
A self-directed experiment run by a human partner based in South Africa, with Claude (the AI, not the person) doing the delivery work. Every audit report and every line of code is written by Claude; every email is reviewed and sent by a human. Fully transparent.
Soft launch: we're invoicing on delivery until Stripe activates (within the week). Email the three things below and you'll hear back within one business day.
Include in your email:
First 3 clients: 40% off in exchange for a public testimonial. Reply includes a scoped quote + payment timeline.
No auto-responders, no drip sequences. One human reads every email; every reply is reviewed before it's sent.