Source: https://oguzerol.com/en/what-to-look-for-when-hiring-a-claude-expert/ [ View full size](https://oguzerol.com/gorseller/senaryo/what-to-look-for-when-hiring-a-claude-expert-senaryo-en.webp) # What to Look for When Hiring a Claude Expert [Oğuz Erol](https://oguzerol.com/en/about-me/) | 5 min read | Published: 4 August 2026 | Updated: 13 October 2026 ## Quick summary What to ask when hiring a Claude expert: questions that separate real Skills, Claude Code and MCP experience, red flags, and a small trial task. **What separates a Claude expert from someone who uses Claude in a chat window is that they can describe a system they built in production, its limits, and a mistake they made.** A CV line saying “I know Claude, ChatGPT, Gemini” doesn’t show that. Below are four parts: which jobs genuinely need Claude-specific expertise, which questions to ask, how to read the answers, and a small trial you can run before deciding. ## First: does your job need Claude-specific expertise? Not every job does. For a blog post, a social media caption or a translation, which model is used often doesn’t matter; a general AI consultant is enough. Claude-specific knowledge makes a difference in work like this: - **Automation with Claude Code:** an agent with file-system and command-line access carrying out a multi-step job under supervision. Which commands are allowed, what happens on error, and at which step a human approves are designed upfront. - **Claude Skills:** turning a recurring job (a checklist, a report layout) into packaged instructions. Details in [What Is Claude Skills](https://oguzerol.com/en/what-is-claude-skills-what-is-it-for/). - **MCP integration:** MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems; it lets a model reach email, a calendar, a database or an ad account ([modelcontextprotocol.io](https://modelcontextprotocol.io/docs/getting-started/intro), as of October 2026). - **Projects and context management:** setups where a team works from shared context. General background on [Claude](https://oguzerol.com/en/what-is-claude-ai-what-does-it-cost-how-to-use-it-from-turkey/) is in a separate article. Rough split: if the job is “get me good text”, a general consultant; if it’s “connect a system to Claude and make it run safely”, Claude-specific expertise. ## Which questions to ask A general question (“do you have AI experience?”) gets a general answer. Ask these, with what you want to hear next to each: | Question | What a good answer looks like | Red flag | | --- | --- | --- | | Where have you used Skills? Can you show an example? | Names a specific job, its steps and what the Skill changed; can show it. | “Yes, I use it a lot, it’s a great tool.” | | Have you built an automation with Claude Code? What limits did you keep commands within, and what happens on error? | Describes an allow-list, approval points and a way to roll back. | Never mentions limits or failure cases. | | Have you connected an MCP server? To which system, with what permissions? | Names the system, the access granted (read-only or write) and why. | “It connects to anything”; no distinction between permissions. | | Before installing a third-party Skill or MCP server, what do you do? | Reads the contents, checks which commands and access it asks for. | “If it’s popular, I install it.” | | What did you decide not to do with Claude, and why? | Can say when a general tool, or no AI at all, was the better choice. | Thinks everything can be done with Claude. | Questions two, three and four are the security side of the job. Giving an agent account access is a serious decision; don’t start with someone who can’t tell you how they will limit that access. This is covered separately in [is it safe to give an AI agent account access](https://oguzerol.com/en/is-it-safe-to-give-an-ai-agent-account-access/). Real experience usually includes a limitation, a bug or an unexpected behaviour, and how it was worked around. Someone with no concrete story to tell has probably not used the tool in production. ## A small trial before you decide If conversation isn’t enough, give a small, paid, reversible trial job: - Pick a single recurring task (for example a report prepared in the same format every week). - Limit access to **read-only**; don’t grant write access in the first trial. - Check the output yourself and watch how the candidate handles mistakes. - Verify which account they accessed with which role, and that access was removed when the job ended. A candidate who tells you when they made a mistake is a better sign than one who shows complete work and never mentions any uncertainty. ## When you receive a proposal - **Scope in writing:** which systems will be accessed, which outputs delivered, when it ends. - **Access stays on your account:** a role is granted instead of sharing a password, and removed when the job is done. - **Time and price depend on the job.** A simple Skill setup and a multi-system MCP setup are not the same job; a proposal promising a fixed duration may not have understood the scope. - **Who maintains it:** tools change often, so who updates the setup afterwards should be settled upfront. ## You can ask me for this work If you need Claude Skills setup, automation with Claude Code or MCP integration, see the [AI Consulting](https://oguzerol.com/en/ai-consulting/) page for details. ## Frequently Asked Questions ### Is a “Claude expert” the same as a “prompt engineer”? No. Prompt engineering is the skill of writing prompts to get good output from a model, and it can work independent of the model. Claude expertise includes that but also covers building automations with Claude Code, packaging Skills, and connecting systems with MCP. ### Can someone without Claude Code experience handle a Claude-related task? For content production or one-off analysis, yes. If the job involves file-system access, running commands or a multi-step automation, an inexperienced person can make mistakes and set access limits wrongly; for those jobs, choose someone who can answer the questions above clearly. ### Why should I ask about MCP? Asking whether someone has connected an MCP server is a fast way to tell “I do integrations” from having actually built one. What you really want to look at is which permissions they granted. ### How long does this take, and how is it billed? It depends on scope. A simple Skill and a multi-system integration don’t take the same time. Ask for a written scope and a list of deliverables; time and price are discussed on that basis. [Back to top ↑](https://oguzerol.com/en/what-to-look-for-when-hiring-a-claude-expert/#icerik) ## You may also find these useful [My Agentic Stack: Claude Code + Skills + MCP, the One-Person EditionI walk through my agentic stack — Claude Code, Skills, MCP connections, and human approval: the setup, the daily flow, the limits, and an honest take.](https://oguzerol.com/en/my-agentic-stack/)[Claude or ChatGPT: Which One for Business Use?Claude or ChatGPT? The differences that matter for work: writing, long documents, code, images, Turkish, plan prices and data settings, plus which I use when.](https://oguzerol.com/en/claude-or-chatgpt/)[Claude Skills I Actually Use as a ConsultantWeekly Google Ads reports, GA4 summaries, SEO tone control, and backup discipline — four real Claude Skills examples I actually use in my consulting work.](https://oguzerol.com/en/claude-skills-i-actually-use-as-a-consultant/) [Browse all articles →](https://oguzerol.com/en/articles/)