Home / Hire engineers / Claude developer

A Claude developer builds and ships on the Claude API and Claude Code

Once Claude is chosen and the design agreed, what is left is code: API calls that return data your product can use, agents on the Claude Agent SDK, MCP servers for your own API, and a Claude Code set-up your repository is ready for. A Claude developer writes that code, ships it and keeps it inside your rate limits. That is the role we staff.

Request Claude developer profiles See every role

Profiles in three daysYou interview and chooseClaude Code Partner

When to hire a Claude developer, and what they ship first.

A Claude developer is hired once Claude is chosen and something has to ship. Each situation lists the code the developer writes first, in your repository and release process.

01
A Claude feature inside your product
Someone sketched the feature in the Claude Console, and now it has to live in your codebase.
The work
- JSON schemas for every answer code reads
- PDFs sent as document blocks
- Citations returned beside each answer
02
An agent on the Claude Agent SDK
A job needs Claude to read files, run commands and call your tools, on servers you control.
The work
- Hooks that block commands outside the task
- Subagents for work that floods context
- Permissions set for each built-in tool
03
Your product needs an MCP server
Customers want to reach your product from Claude and their own agents, and you have an API but no server.
The work
- Tools shaped around user tasks, not endpoints
- Descriptions tested on real requests
- Local and remote versions of the server
04
Claude Code set up per laptop
Each developer keeps their own instructions and permissions, so one repository gets different treatment in every session.
The work
- One CLAUDE.md checked into the repository
- Hooks that format and type-check edits
- Shared skills for routine tasks
05
Routine fixes waiting in the backlog
Small, well-described issues sit for weeks because nobody has an afternoon to spare for them.
The work
- Claude Code running in GitHub Actions
- Runs started only by a label
- Allowed tools set in the workflow file
06
A prototype that breaks under real traffic
It worked in the demo. Under load it hits rate limits and overloads, and long answers stop mid-sentence.
The work
- Retries that wait as the API asks
- Every stop reason handled in code
- Peak traffic checked against rate limits

One Claude change, from ticket to production.

The life of one change in five stages. The Claude-specific work sits in the calls, the responses and the limits they run under, which is where a general backend developer usually gets caught out.

01
Plan
Passed
Claude Code reads the ticket and the code it touches and proposes a plan, which the developer corrects before any file changes.
Claude Code, plan mode
02
API calls
Passed
Messages API requests with tool definitions, the right model and effort for the task, and a JSON schema for every response the code reads.
Messages API, tool use, structured outputs
03
Response handling
Passed
Code for every stop reason: tool calls run, refusals shown as refusals, and responses cut off at the token limit continued or discarded, never parsed.
Stop reasons, retries, schemas
04
Review
Waiting
Opened from Claude Code with test results and the plan attached. A senior Claude developer reads every line, and the model choice, before it merges.
Pull request, CI checks
05
Limits and cache
Waiting
Peak traffic checked against your rate and spend limits before switch-on, then cache reads tracked against cache writes per feature, so an edit that breaks caching shows within days.
Rate limits, usage fields

Toolkits a Claude developer works in.

The calls and responses are covered stage by stage above. Around them, a Claude developer keeps six toolkits in shape, from the SDK your service uses to tests that never call the API.

Anthropic SDKs

The official SDK for the language your service already runs, so Claude arrives as a dependency rather than as a new service.

Python SDK
TypeScript SDK
Java SDK
Go SDK

MCP servers

Servers for your own API, checked in the MCP Inspector before any agent or customer connects to them.

MCP TypeScript SDK
MCP Python SDK
MCP Inspector

Repository set-up

The CLAUDE.md file, settings, hooks and skills kept in version control, so every developer’s sessions follow the same rules.

CLAUDE.md
Hooks
Skills

CI workflows

Claude Code started from a label or a comment in GitHub Actions, with the tools it may use fixed in the workflow file.

claude-code-action
GitHub Actions

Batches and files

Backfills and nightly re-runs sent as Message Batches at half the token price, and large files uploaded once, then reused by ID.

Message Batches
Files API

Offline tests

Recorded Claude responses and stubbed tools, so the suite runs in CI on every pull request without spending a token.

pytest
Vitest

Seniority, and what each level owns

Each level is set by the largest change it makes alone: a tool handler, the model choice for a feature, or the rules every repository follows.

Mid · 3–5 years: Takes a scoped ticket from plan to merge: the API calls, the tool handlers and a test for each stop reason. Works inside the repository’s Claude Code rules rather than writing new ones. A change of model, or anything that raises the cost per request, goes to a senior developer.
Senior · 5–8 years: Owns a Claude feature: picks the model and effort for each call on measured results, designs the tool schemas and the MCP server around your API, and says yes or no to any change that raises cost per request. Reads every agent-written pull request before it merges.
Lead · 8+ years: Owns the Claude Code set-up every repository starts from, from the settings file and hooks to the review checklist, and the CI workflows that let Claude Code open pull requests. Decides with your CTO which work agents may start on their own and which waits for a label.

AI-assisted Claude development, with a senior developer answering for every change.

A Claude developer writes much of the code in Claude Code, then ships features that call Claude, so one set of repository rules governs both the agent and what it builds. The decisions that are expensive to reverse stay with the developer: the model for each call, the tool schemas and anything that changes cost.

Change set · ticket triage on ClaudeDrafted with an agent

Example · Claude Code session

Added Triage output as a JSON schemano prose parsing
Added Refusal and max_tokens handlingno silent failure
Changed Client timeout and max_retries setSDK backs off
Removed Hand-written JSON repairschema enforces it
Changed Model for triage: Sonnet → OpusHuman decision

Four changes ready; model switch held

Senior Claude developerCompared both models on the same ticketsKept on Sonnet

Revert model, then merge

Calls tested in the session that writes them

Claude Code runs the feature’s own API calls against sample inputs while it writes them, so a schema or stop-reason bug shows before the pull request.

The plan is part of the review

A pull request from Claude Code carries the plan it followed, so a senior Claude developer can send back a wrong approach before reading the diff.

Allowed commands set in the repository

The commands Claude Code may run without asking sit in the repository’s settings, and a deny rule blocks pushing, publishing and deleting, so nobody approves them by reflex.

From your requirements to matching developers in 3 working days.

You do not need to enter a long recruitment process before you know whether we have the right people.

Tell us what you need

Send the role, stack, seniority, required experience, working-hour overlap, team size, and expected start date.

Get matching profiles

Within three working days, we send relevant profiles with CVs, project experience, English level, availability, and rates.

Interview the developers yourself

Use your own technical interview and selection process. Meet the people who would actually join the team and reject anyone you are not confident about.

Start working together

Once you choose, we handle the contract and onboarding. Your developers join your repositories, standups, tooling, and delivery process.

Choose the engagement model that fits the gap.

You can hire software developers through an outsourcing model that matches how much ownership, capacity, and flexibility your team needs. Most clients start with one engineer and grow. Roles and stacks are further down the page.

Ad-hoc staff augmentation

One engineer or several, embedded in your team and managed by you. The most common way clients start with us.

Best for a named capability or capacity gap you already understand.

Speed: Profiles in 3 days
Pricing: Per hour
Managed by: You

Dedicated team

A complete team for a product, a module or a roadmap stream — engineering, QA, DevOps and design as the work needs it.

Best for stable delivery capacity without hiring every role in-house.

Speed: 2–4 weeks to assemble
Pricing: Monthly per person
Managed by: You, or jointly with our lead

Recruitment

We recruit for a skill set we do not have on the bench, against your requirements, and the person joins on your contract.

Best for a niche stack, a rare seniority level, or a role you keep failing to fill locally.

Speed: 2–4 weeks
Pricing: Per hire
Managed by: You

Build, operate, transfer

We hire and run the team in Poland or Ukraine, then hand it over to you as your own staff for direct contracts when you are ready.

Best for opening a development centre in Europe without setting one up yourself first.

Speed: 3-4 weeks to build
Pricing: Monthly per person, plus a transfer fee
Managed by: Us, then you

The reason to outstaff is speed. The reason to stay is quality.

Netguru, a 500-person Polish consultancy whose own clients include UBS, has staffed projects with our engineers for more than 8 years. They could hire anyone.

Staff augmentationNetguru, 2018-ongoing

Read the Netguru case study

8+years

Netguru has staffed projects with our engineers

500+people

at Netguru, whose own clients include UBS

3days

From your requirements to matching profiles with CVs

90%

Of our developers have 5 to 20 years of experience

6years

Our longest continuous engagement, still running

10+

Time zones, with working hours flexed to match your team

What it is like to have our engineers on your team.

Questions about hiring a Claude developer.

How is a Claude developer different from an AI engineer?

In depth on one platform. An AI engineer connects whichever model you use to your systems and data. A Claude developer is hired once Claude is chosen, for fluency in Claude’s own surface: tool use, prompt caching and batches on its API, the Agent SDK, MCP servers and Claude Code. If the model is still an open question, that decision comes first.

Can the developer set Claude Code up for our team?

For the repositories they work in, yes: the CLAUDE.md file, hooks that run your linters and tests, shared skills and a GitHub Actions workflow, all committed so every session follows them. Seats, billing and organisation-wide settings stay with whoever administers your Claude plan.

Does it matter which language our backend is in?

Rarely. Anthropic publishes SDKs for Python, TypeScript, Java, Go, Ruby, C# and PHP, and the developer works in the one your service is written in, adding Claude as a dependency rather than standing up a separate service for it. MCP has official SDKs in most of the same languages, so a server can live beside your API.

What happens when Claude refuses, runs out of tokens or is overloaded?

Each case has its own path in the code. A refusal is shown as a refusal, not retried as if nothing happened. A response cut off at the token limit is continued or discarded, never parsed. Rate-limit and overload errors are retried after the wait the API asks for, and repeated failures alert a person instead of looping.

Can the developer build an MCP server for our product?

Yes. The server wraps your existing API with one tool per task a user would ask for, not one per endpoint, and each tool description is tried against real requests before release. It can run locally for your own developers or remotely for customers, with the same permissions your API already enforces.

Is a Claude developer the same as a Claude Certified Developer?

Not necessarily. Claude Certified Developer is Anthropic’s exam for engineers who build with the Claude API, Claude Code and MCP, from a first integration to production agents. This page uses the term for the job, not the credential, and does not claim it for our engineers. Our checks cover the same ground: candidates show code they shipped on an LLM API and explain a failure it had to survive.

Does Claude Code send our source code to Anthropic?

It sends the files and command output it reads to the model, under your organisation’s agreement with Anthropic or your cloud provider, never a personal account. The repository’s settings decide which paths and commands it may touch, and files such as .env can be denied outright. Whether that data may be used for training is set by your commercial terms, so check them with your legal team.

Which API key does the developer use?

One your team issues for a development workspace in your Claude organisation, never a personal key. Its usage and spend show under that workspace, apart from production, and it can be revoked in one step. Production keys stay with your team or your secrets manager, and the developer never needs to read them.

Can we start with one senior developer?

Yes. Many clients start with one senior engineer to cover a specific capability or capacity gap, then add specialists only when the roadmap requires them.

Can we scale the team up or down?

Yes. Adding an engineer follows the same route as the first, and you interview them. Reducing the team follows the notice period agreed in the contract, and the work is handed over in your repository before anyone leaves.

Will the developer work as part of our team?

Yes. You manage priorities and day-to-day work directly, in your own channels and delivery process.

Do we interview the developer before they start?

Yes. You interview every candidate using your own process and make the final decision about who joins the team.

Are the developers your employees or freelancers?

Our engineers are ImproveIT employees and long-term contractors working through our teams in Poland and Ukraine. We are not a freelance marketplace and do not resell developers from other vendors.

What if the developer is not the right fit?

Replacement terms are defined in the contract upfront, so it does not become a negotiation.

Who owns the code and intellectual property?

You do. IP assignment is included in the contract, and we can sign an NDA before discussing the project.

How do you handle security, access and NDAs?

We sign an NDA before discussing the project, and confidentiality and IP assignment are part of the contract. Engineers work in your repositories, cloud accounts and tools with the access your team grants, so you can narrow or revoke it at any time.

Can your developers work with our timezone?

Yes. Our engineers work with teams across Europe and the US. We agree the required working-hour overlap before matching candidates.

How is their English?

Our developers work in English with international product and engineering teams every day. English level is included in the candidate information you receive before the interview.

What does it cost to hire a developer?

Rates depend on the role, seniority, experience, and engagement model. You receive the rate, availability, and candidate profile in writing before making any commitment.

What are the minimum engagement and notice period?

Both are agreed before the engagement starts and included in the proposal and contract. We are flexible in that and usually agree on the terms suitable for a client.

Taking the feature further?

A design settled before more gets built, agents delivered as a project, or an assistant your staff can use as it is.

Tell us which engineers you need.

Stack, seniority, how many and when. We come back with matching profiles, availability and rates within three working days.

You interview them with your own process, and you reject anyone you are not sure about.

1Send the requirements
2Get profiles with CVs, availability and rates in 3 days
3Interview, choose, and start within a week

Book a call instead

We sign an NDA before discussing anything in detail. We reply within 24 hours.

A Claude developer builds and ships on the Claude API and Claude Code

When to hire a Claude developer, and what they ship first.

A Claude feature inside your product

An agent on the Claude Agent SDK

Your product needs an MCP server

Claude Code set up per laptop

Routine fixes waiting in the backlog

A prototype that breaks under real traffic