A Claude Code consultant sets up the files that tell Claude Code how to work in your repository: project instructions, custom subagents, skills, MCP server connections, hooks and permission settings. At handoff, those files stay in your repo, and your team can read, change and version them like any other code.
That gives you a way to scope the job before you hire. Ask what files you get, not how many hours you buy. The table below lists each deliverable, where Anthropic's documentation puts it, what the docs say it controls, and what to ask about it.
What a Claude Code setup changes
Anthropic describes Claude Code as "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools" (Claude Code overview). It runs in the terminal, in IDE extensions, in a desktop app and on the web.
The overview page also states that every surface connects to the same engine. A repo's CLAUDE.md files, settings and MCP servers work across all of them. For a buyer, that means one setup in the repo serves every developer on the team, whichever surface each person uses.
Out of the box, Claude Code knows nothing about your build commands, your conventions or the systems it must not touch. A consultant writes that knowledge into files. The quality of the engagement is the quality of those files.
The deliverables table
Each row is one artifact. Every "what the docs say" cell links to the Claude Code docs page it comes from.
| Deliverable | Where it lives in your repo | What the docs say it controls | What to ask the consultant |
| Project instructions | CLAUDE.md or .claude/CLAUDE.md at the project root | Instructions Claude reads at the start of every session, such as build and test commands, coding standards and architecture decisions. They reach the team through version control. The docs set a target of under 200 lines per file. (memory) | How many lines is it, and which lines came from our own standards? |
| Path-scoped rules | .claude/rules/, one markdown file per topic | Modular instruction files. A rule can be scoped to file paths, so it loads only when Claude works with matching files. (memory) | Which rules load on every task, and which load only for certain folders? |
| Custom subagents | .claude/agents/ | Specialized assistants. Each one runs in its own context window, with a custom system prompt, specific tool access and independent permissions. Frontmatter fields set its tools, denied tools, model and permission mode. (subagents) | What job does each subagent do, and why does it need each tool it has? |
| Skills | .claude/skills/, one folder per skill with a SKILL.md file | Packaged instructions. Claude uses a skill when it is relevant, or a person runs it with a slash command. The description field tells Claude when to apply it. (skills) | Which skills run on their own, and which run only on command? |
| MCP server connections | .mcp.json at the project root | Connections that give Claude Code access to your tools, databases and APIs. Project-scoped servers reach the team through version control. Claude Code asks for approval before it uses them in an interactive session. (MCP) | Which servers read outside content, and who maintains each one? |
| Hooks | .claude/settings.json, or a skill's or subagent's frontmatter | Shell commands, HTTP endpoints, MCP tool calls, prompts or subagents that run at set points in Claude Code's lifecycle. A PreToolUse hook can block a tool call. (hooks) | Which hooks block actions, and what exactly does each one block? |
| Shared settings and permissions | .claude/settings.json | Team permissions, hooks, plugins and the environment variables the project needs. Allow, ask and deny rules decide what Claude Code may run without a prompt. (settings) | Which commands run without a prompt, and which ones are denied? |
Check that every file is in project scope
The docs give most of these artifacts two homes: one in the project, and one in a personal folder. Only the project copy reaches your team.
- Instructions: CLAUDE.md is shared through version control. CLAUDE.local.md is personal, and the docs say to add it to .gitignore (memory).
- Subagents: files in .claude/agents/ belong to the project. Files in ~/.claude/agents/ apply to one person's projects on one machine (subagents).
- Skills: the docs say to commit a project skill "so your team gets it too." A skill in ~/.claude/skills/ stays on one machine (skills).
- MCP servers: project scope writes .mcp.json. Local and user scope write to ~/.claude.json, which stays private to one user (MCP).
- Settings: .claude/settings.json is the shared file. .claude/settings.local.json is personal, and Claude Code keeps it out of git when it creates the file (settings).
This split is the main handoff risk. A consultant can build a setup that works well on their own machine and still leave you nothing. Before the final invoice, clone the repo on a clean machine and confirm that the setup still loads.
Instructions guide Claude, hooks and deny rules enforce
The memory docs state that Claude treats CLAUDE.md "as context, not enforced configuration." To block an action whatever Claude decides, the docs point to a PreToolUse hook instead (memory).
This line separates a careful setup from a fragile one. Suppose your requirement is "never edit the production config." A sentence in CLAUDE.md asks Claude to comply. A deny rule or a blocking hook stops the action.
The settings page adds a detail that matters for teams. Deny and ask rules in the shared settings file apply right away. Allow rules wait until each teammate trusts the folder (settings). So a test can pass in a folder the consultant already trusts and still prompt a new teammate.
Ask the consultant to sort every hard requirement into one of three places: an instruction, a permission rule or a hook. Anything that must never happen belongs in the last two.
Questions to ask before you hire
- Which files will you add, and at which paths? Expect the project paths from the table, not files in a home directory.
- Can our team change each file without you? Skills are markdown files with YAML frontmatter, per the skills docs. Settings files are strict JSON, per the settings docs. Your developers can edit both.
- Which of our rules are hard limits? Expect hooks or deny rules for those, not CLAUDE.md lines.
- Which MCP servers will you connect, and why? The MCP docs say to verify you trust each server, because servers that fetch external content can expose you to prompt injection (MCP).
- How long are the instruction files? The docs say longer files consume more context and reduce adherence (memory).
- Do you keep one instruction file for every coding agent? If your repo already has an AGENTS.md and no CLAUDE.md, Claude Code reads the AGENTS.md as its project instructions (memory). A second file that repeats it is a maintenance cost.
- How will we check the handoff? The memory docs describe a prompt audit that finds outdated or conflicting instructions across CLAUDE.md, rules, skills and subagents. Ask the consultant to run it with your team on the last day.
What the directory can and cannot tell you
This directory has no Claude Code tag. Listings tag the tools a provider works with, and Claude is one tag. So this site cannot count Claude Code consultants.
The nearest real figures come from the directory's own data. Of the 171 listed providers, 76 tag Claude. Of those 76, 73 list workflow automation and 66 list custom AI builds. By price band, 29 are budget, 42 are mid-range and 5 are premium.
Those numbers say who works with Claude. They do not say who sets up Claude Code. Use the questions above to find out.
Where to find providers who work on Claude
Start with the providers who work on Claude. If you run an agency, Claude providers that serve agencies narrows the list. For a build that goes past repo setup, such as an internal agent or a custom model workflow, see the custom AI providers. If your team wants a plan before any build, the AI strategy providers list covers that work.
Send each shortlisted provider the deliverables table and the seven questions. A provider who answers in file paths and permission rules can scope a Claude Code engagement. A provider who answers only in hours cannot yet.
For context on how teams use Claude Code after the setup is done, read what 400,000 Claude Code sessions reveal about agentic coding.