Cookies on AIGROW

We use necessary cookies to run the site. With your permission, we also use cookies for analytics, session recordings and marketing attribution.

Cookie Policy

API documentation tested by coding agents: 20 integration tasks attempted from your docs alone, with the gaps that stopped the agent and the passages rewritten

$990Fixed price, paid onceReport within 7 working days of your docs and sandbox key

Order for $990How it works

We work only with a sandbox or test key, never production access. If we do not deliver what the page lists, you get a full refund.

AIGROW tests your API documentation with a coding agent for $990, paid once. Claude Code attempts 20 integration tasks, such as authentication, pagination or webhooks, in a clean sandbox with only your docs and your sandbox API. Within 7 working days you get every result, where the agent got stuck, the 10 costliest doc gaps and rewritten passages.

What the test does, and what it does not

A coding agent integrating your API has only your docs to go on and nobody to ask. The test gives it real integration tasks and records where it stops or guesses.

What it does
  • Writes 20 typical integration tasks from your docs, such as authenticating, a first call, pagination, webhooks, errors and retries, installing your SDK and a common use case, or uses your own task list
  • Runs Claude Code on each task in a fresh container that can reach only your docs, your sandbox API and the package registry, with web search turned off
  • Gives the agent your llms.txt as well, when you have one
  • Allows up to 3 attempts per task with the same fixed prompt, and checks each result against your sandbox API
  • Records for each task whether it passed, the attempts it took, where the agent got stuck, the endpoints or parameters it invented and the doc page responsible
  • Ranks the 10 doc gaps that cost the most attempts and rewrites the passages behind them, ready to paste
  • Writes a suggested llms.txt and llms-full.txt from your docs if you have none
  • On request, runs the same tasks with a second coding agent from another vendor, such as OpenAI Codex, and sets the two side by side
What it does not do
  • It does not use production credentials: only a sandbox or test key with a spend limit, which we delete when the work ends
  • It does not edit your docs site. We hand over the rewritten passages and your team publishes them
  • It does not promise more signups or adoption, or that any AI assistant will read your docs
  • It does not load test or security test your API. Where the API behaves differently from its docs, the report says so
  • It does not include a second run after you fix the docs: a rerun of the same tasks is a new order at the same price

From payment to the report

Each step and when it happens. The 7 working days start when we have your docs and a sandbox key that works.

Pay at checkoutPay by card at the Stripe checkout. Stripe sends the receipt.2 minutes
Send your docs and a sandbox keyWe email you straight after payment. Reply with your docs, the languages to cover and any tasks of your own, and send the sandbox key in a separate reply.About 20 minutes
We write the tasksWe write 20 tasks from your docs, or take yours, each with the check it must pass, and send you the list before any run starts.Within 2 working daysOf your docs and key
The agent runs each taskEach attempt starts in a fresh container that reaches only your docs, your sandbox API and the package registry. The agent gets the same prompt every time.Up to 3 attempts per task
The reportYou get the task table, the 10 most costly gaps with the rewritten passages, a suggested llms.txt if you have none, and every log.Within 7 working daysOf your docs and a working sandbox key
The key is deletedWe delete the sandbox key on our side and confirm it in writing. Revoke it on your side too.At delivery

One test of your docs, one price

Paid once at checkout, which asks for your email and card. Right after payment we email you a few written questions. The report follows within 7 working days of your docs and a working sandbox key.

What we need from you

Reply to our email with the address of your public docs, or an export of private ones, and the languages or SDKs to cover. Add your own task list if you have one. Send a sandbox or test key with a spend limit and no production access in a separate reply.

What you receive

  • The 20 tasks, each with its starting point and the check it must pass
  • A task table: passed or failed, attempts used, where the agent got stuck, invented endpoints or parameters, and the doc page responsible
  • The 10 most costly doc gaps, ranked by the attempts they cost
  • Rewritten doc passages for those gaps, ready to paste
  • A suggested llms.txt and llms-full.txt, if you have none
  • The full log of every attempt

API documentation tested by coding agents

$990

Paid once at checkout. No subscription, nothing renews.

If we do not deliver what this page lists, or you change your mind before work starts, you get a full refund.

Why buy it from AIGROW

What you can check before you pay, and after the work.

A fixed price
$990, paid once. No subscription, nothing renews. Everything included is listed on this page before you pay.
Proof you can check
Every result in the report links to the full log of the attempt behind it and names the doc page the agent was reading, so your team can follow what it tried and check each finding.
A public method
The free llms.txt validator checks the structure of your llms.txt and tests up to 20 of its links. When you have the file, the agent in the test reads it too. Run the llms.txt validator

Before you buy

Who it is for, the agent we use, the key we need, llms.txt, reruns and what we do not promise.

Developer relations and documentation teams at companies with a public API, SDK or developer tool, who want to know whether a coding agent can integrate their product from the docs alone.

Claude Code, Anthropic’s coding agent, run without a person in the loop and with the same fixed prompt for every task. If you want a comparison, say so in your reply: the same tasks then run with a second coding agent from another vendor, such as OpenAI Codex, and the report shows both side by side.

A sandbox or test key with a spend limit and no access to production data or money. Never send production credentials: if a key turns out to reach production, we stop, tell you to revoke it and ask for a sandbox key. You can revoke the key at any time, and we delete it when the work ends.

The agent calls your API to check its own work, so we need a sandbox or test account before the runs start. If you cannot give us one, ask for a full refund before the runs start.

Each task has a check written before the runs, such as a script that asks your sandbox whether the record exists or the webhook arrived. A task passes when its check passes within 3 attempts. The table also counts the attempts, so a pass on the third try still shows the gap.

No. llms.txt is a proposal from September 2024 for a file that points language models to the main pages of a site. If you have one, the agent gets it with your docs; if you have none, we write a suggested llms.txt and llms-full.txt from your docs. We do not claim that any AI assistant reads the file on its own.

We do not promise that, or more signups. The test measures how a coding agent does on your tasks with your docs, and the rewritten passages remove what stopped it. What developers choose is up to them.

Yes. Order the test again at the same price and say it is a rerun: we run the same tasks with the same prompt and show the results next to the first run.

Yes. Send an export, such as Markdown or HTML files or an OpenAPI file, and the agent reads that copy in place of your public site.

They are used for your report only. The key is deleted at delivery, and your files and logs 30 days after delivery, or sooner if you ask, with a written confirmation. Claude Code runs under Anthropic’s commercial terms, which say Anthropic does not train its models on the code or prompts sent to it.

No. If your own product calls OpenAI, Anthropic or Google models and the bill keeps growing, the LLM API cost review, at $990, breaks the spend down by feature and writes the changes that cut it.

Can AI crawlers reach your docs?

The test hands the agent your docs directly. Whether AI crawlers can reach them on the open web is a separate check: the AI crawler access setup reads your robots.txt, CDN rules and sitemap, and sets them for the crawlers you choose.

See the AI crawler access setup

The service that fits next

$2,900

MCP server for your business

We build and host a read-only MCP server over your catalog, stock, prices, hours or docs, so Claude, ChatGPT and other AI agents can look them up. Hosting is billed monthly from handover.

See the MCP server service

Find where a coding agent gets lost in your docs.

Pay once, send your docs and a sandbox key, and get the task results, the 10 most costly gaps and the rewritten passages within 7 working days.

Order for $990See every done-for-you service

$990 · paid once · delivered as listed or refunded