Claude API integration that puts Claude in your product, with costs and risks undercontrol.
Need a Claude AI developer? We connect Anthropic’s Claude models to your app, website or internal tool through the API. We choose the model for the job, test the prompts on your real examples, keep the key on your server and cap what it can spend.
Claude can do the task. Can your app handle the rest?
Businesses hand whole tasks to Claude through the API. Anthropic's own research and documentation show what that asks of the app around it: long prompts, a short cache and models that are retired on a schedule.
77%
of the business API conversations Anthropic sampled in August 2025 showed automation, where the task is handed over to Claude
What we doWe agree with you which steps Claude may finish alone, and risky actions wait for a person to approve them.
What a Claude integration needs before customers use it.
Five problems owners bring to us about building with the Claude API, and what we do about each.
Problem 01 · Key in the browser
Our first version calls Claude from the web page, so anyone who opens the page can copy our API key.
What we do
We move the call to your own server. The page talks to your server, and your server holds the key and talks to Claude. Anthropic's documentation recommends this set-up. We also give test and live their own keys and workspaces in your Claude Console.
The key held on your server only
A server route between page and Claude
Separate keys for test and live
Old, exposed keys deleted
Sign-in and per-user limits on your route
Problem 02 · Rising cost
We send the same long instructions and documents with every request, and the bill grows faster than our usage.
What we do
We count the tokens in each request and cut what the question does not need. The part that repeats goes first, so Claude's prompt cache can reuse it. Jobs that can wait are sent in batches. A monthly spend limit is set in your Console.
Tokens counted before sending
Only the pages a question needs
The repeated part cached
Batches for work that can wait
A spend limit in your Console
Problem 03 · Answers to check
Claude sounds sure of itself even when it is wrong, and we have no way to tell before a customer sees it.
What we do
We test every prompt on real examples from your business and you read the results. Where the answer feeds your software, we ask for it in a fixed structure and check it in code. For answers drawn from your documents, we show the passage they came from.
Real examples, awkward ones included
Results you can read before launch
Structured answers checked in code
The source passage shown
The same tests run after each change
Problem 04 · Tools that act
We want Claude to look up orders and update records, but we worry about what it could do by mistake.
What we do
With tool use, Claude asks for an action and your own code carries it out. That gives us the place to add rules. We define each tool narrowly, check what Claude passes to it, and make payments, deletions and messages to customers wait for a person.
Each tool does one narrow thing
Inputs checked against a strict schema
Read-only tools where they are enough
Risky actions wait for approval
A cap on steps in one task
Problem 05 · Limits and outages
On a busy day the API returned errors, and our feature showed customers a blank screen.
What we do
Claude's API limits requests and tokens per minute and can report that it is overloaded. We read the wait time it sends back, retry with a pause and stream long answers. If it still fails, your app falls back to a queue, a simpler path or a clear message.
Rate-limit replies read and obeyed
Retries with a growing pause
Long answers streamed
A fallback your customers can live with
An alert when errors rise
Two reports from Anthropic. Two lessons for your build.
Anthropic published both of these itself. They show what can happen on the model's side of an integration. Neither involves a client of ours.
01
Published postmortem · September 2025
Three faults lowered Claude's answer quality for weeks
What happened
Anthropic reported that three infrastructure bugs intermittently degraded Claude's responses between August and early September 2025. At the worst hour, 16% of requests to one model were affected by the first bug. Anthropic said its standard evaluations did not catch the problem, and that it would run more sensitive checks on its live systems.
What we do about it
We keep your test examples and run them again on a schedule and after each change, so a drop in answer quality shows up in your own results.
Anthropic and Andon Labs let a Claude model run a small shop in Anthropic's office for about a month. It could search the web, email suppliers, change prices and chat with customers. It was talked into discount codes and free items, priced some goods below cost and told customers to pay into an account it had made up. The shop did not make money.
What we do about it
We define each tool narrowly, check what Claude passes to it and make prices, payments and messages to customers wait for a person.
Live websites we built. Open them and judge the work for yourself. They are not Claude integrations: we don't say which model provider a client's product uses, and private work is shared on a call with the client's agreement.
Why hire Sun Software as your Claude AI developer?
We build with Claude, and with MCP, the protocol used to connect AI models to tools.
Claude is one of the models our team builds with, and we build with MCP, the Model Context Protocol. This website runs a small read-only MCP server of its own. We test every AI system on real examples before it goes live, and we cap what it can spend. We have no partnership with Anthropic, and we check its documentation on the day before we rely on a feature or a limit.
The Sun Software team
01
Tested before it is built in
Your real examples are run first, and you read the answers before launch.
02
Spend is capped in two places
A limit in your Claude Console, and limits per user in your own app.
03
Actions wait for a person
Claude may ask for a payment, a deletion or a customer message. A person approves it.
04
Checked against the documentation
Features, limits and retirement dates are read on Anthropic's own pages, with the date noted.
Great....very professional, delivered on time and exactly as I needed. Exceptional. I always find Sunsoft very easy to work with, always helpful and I'll continue using their services for the foreseeable future. No need to go anywhere else.
Client in United KingdomOctober 2026 · Fiverr review
“
Excellent experience from start to finish. Communication was great, responses were always prompt, and the work was completed quickly. They captured the exact look and feel I wanted on the first try and delivered excellent value for the money. I would gladly work with them again and highly recommend their services.
Client in GermanySeptember 2026 · Fiverr review
“
Choosing SunSoft was exactly the right decision. Their professionalism, extensive experience, and attention to detail have really impressed me so far. It was a complex project that exceeded my expectations! I look forward to continuing to work with SunSoft!
Client in CzechiaAugust 2026 · Fiverr review
Questions about building with Claude
What does a Claude AI developer do?
The developer connects Anthropic’s Claude models to your product through the API. With us that means choosing the model for the job, testing the prompts on your real examples, keeping the key on your server, adding tools with approvals and setting spend limits.
Can my website call Claude straight from the browser?
It should not. Anything in a web page can be read by the person using it, and that includes an API key. Anthropic’s own documentation says to call the API from your back end. We put a small server route between your page and Claude, and the key stays there.
What will Claude cost us to run each month?
It depends on how many requests you send and how long they are, because the API is billed by the text going in and the text coming out. You pay Anthropic directly. We estimate the usage from your own volumes during testing, and we set a spend limit in your account so a mistake cannot run on unnoticed.
What does Anthropic do with the data we send?
We don’t answer that from memory, because the terms differ by plan and can change. We read Anthropic’s published data retention and usage pages with you for the plan you will use, and give you the links for your own adviser. We agree with you which data may be sent, and build to that.
Can you move our AI feature from another provider to Claude?
Yes. Anthropic offers a compatibility layer for trying Claude with code written for another provider, and its documentation says that is for testing, not for long-term use. So we test with it, then build on the Claude API itself and rework the prompts. Your old and new versions are scored on the same examples.
Are you an Anthropic partner?
No. Claude is one of the models our team builds with, alongside models from other providers. We have no partnership or reseller deal with Anthropic, so if another model suits your job better, we will say so.
What decides the cost of Claude API integration services?
You get a written proposal once we know the job Claude will do in your product. Usage is billed by Anthropic to your own account, and it is separate from our fee. Three things decide it. The job: One task such as summarising a document, or several tasks with different prompts and tests. Tools and actions: Whether Claude only reads and answers, or also looks things up and acts in your systems. Volume and limits: How many requests you expect, how long they are and how much control over spend you need.
What happens after I contact you?
We reply within one business day. Then the work runs in 6 steps: Define the job, Choose the model, Set up the account and keys, Write and test the prompts, Add tools and approvals, Limits, fallback and handover. Each step ends with something you can review.
1Plan My Claude Integration
2We reply within one business day
3Discovery call
4Step one: Define the job
Adding Claude to your product, or fixing a first attempt?
Tell us what you want Claude to do in your product and send a few real examples. We start by testing those examples, before anything is built.