02Getting started
Connect with AINew
Give an assistant our documentation and it will write the integration for you. Every page has a plain-text version, and the whole documentation is available as one file.
How to give an assistant the documentation
Choose the way that suits you. All of them work with any assistant: ChatGPT, Claude, Cursor, and others.
- Copy for AI
- The button above the heading of every page puts its text on the clipboard as Markdown. Paste the text into a chat and the assistant sees the same page you do.
- Open in ChatGPT or Claude
- Two buttons side by side open a chat and pass the page address in
.mdformat right away. - A page as .md
- Add
.mdto the address of any documentation page and you get clean text without the layout, for examplerendergrid.io/.docs/ quickstart.md - llms.txt
- A list of all pages with one line about each. The assistant uses it to choose what to read.
- llms-full.txt
- The whole documentation in one file, for an assistant that needs to know everything at once.
Start with the “Quickstart” page: it is enough for your first image. The assistant can read the rest by following its links.
A ready-made prompt for the assistant
Copy the text, add your task, and send it to the assistant. It already contains the rules that are forgotten most often.
Connect image generation through the Rendergrid API in my project. Documentation: https://rendergrid.io/ llms.txt Full text: https:// rendergrid.io/llms-full.txt Rules: - read the key from the RENDERGRID_API_KEY environment variable; do not put it in the code; - add X-Idempotency-Key to every generation request and save it for retries; - poll the task status no more than once every 5 seconds, or accept a webhook; - save the image from result_urls right away: the link stays valid for a day; - on 429 and 503 responses, wait as many seconds as Retry-After says.
Files the documentation provides
| Address | What is inside | When to use it |
|---|---|---|
/llms.txt | A list of pages with short descriptions | A first look, finding the page you need |
/llms-full.txt | The full text of the documentation | The assistant needs to know everything without following links |
/docs/ | One page as Markdown | A specific question: webhooks, errors, limits |
/openapi.json | A description of all API methods | Generating a client—OpenAPI page |
/skills/ | A skill for Claude Code, in English | You work in Claude Code—“Skills” section |
/skills/ | A rule for Cursor, in English | You work in Cursor—“Skills” section |
The files are updated together with the documentation. The date of the last update is shown on every page, and the list of changes is in the “What’s new” section.
An AGENTS.md file for your project
Put this file in the root of your repository. Many coding assistants read it on their own and follow its rules without being reminded.
MCP server
MCP connects Rendergrid to your assistant as a set of tools: the assistant creates an image, waits for the result, and checks your balance on its own. It works with Claude Desktop, Cursor, and other programs that support MCP. There are two ways to connect; the tools and prices are the same in both.
Remote server. The program connects to us at https:// (also available as https://) and passes your API key in the Authorization: Bearer or X-API-Key header. There is nothing to install. A request without a key gets 401.
Local bridge. If the program cannot connect by address, run the @rendergrid/mcp package through npx; Node.js 22+ is required. The bridge reads the key from an environment variable.
Replace rg_live_… with your key: this file does not expand environment variables.
| Tool | What it does | API method |
|---|---|---|
generate_image | Creates an image and waits for the result for up to 45 seconds | POST /images/ |
get_creation | Returns the status and file links; if the task is still running, waits for up to 45 seconds | GET /creations/ |
upload_image | Uploads a reference image for editing: PNG, JPEG, or WebP up to 25 MiB | POST /uploads |
list_models | Shows the models and prices of your account | GET /models |
get_balance | Shows the available amount |
Skills for Claude Code and Cursor
A skill is a ready-made instruction for an assistant that writes code: how to call the Rendergrid API, which fields are required, how to get the result, and what to do on errors. Download the file and put it in your project—the assistant picks up the rules on its own. The files are written in English: they are read by a model, not by a person.
- Claude Code
- The file
/skills/. Put it inclaude-code/ SKILL.md .claude/skills/rendergrid/SKILL.mdin your project. - Cursor
- The file
/skills/. Put it incursor/ rendergrid.mdc .cursor/rules/rendergrid.mdcin your project.
The skills are built from this documentation and updated along with it. Do not write the key into the file: have the assistant read it from the RENDERGRID_API_KEY environment variable.
What not to give an assistant
- Your API key and webhook secret. Ask the assistant to read them from environment variables, and do not paste the values into the chat.
- Your clients’ data. Use made-up values for examples:
you@example.com,5b0214ae-9e39-460a-820a-f6ae3e24400b. - Blind trust in the code. Check what the assistant wrote against this documentation, especially the idempotency key, the polling interval, and the webhook signature check.
Frequently asked questions
Can I give the assistant a link to the documentation?
rendergrid.io/. If it cannot, click “Copy for AI” and paste the page text into the chat.Is it safe to give the assistant my key?
RENDERGRID_API_KEY environment variable, and do not paste the value into the chat.Is money charged when I work through MCP?
What if the image is not ready within 45 seconds?
generate_image tool returns the task ID and a hint to call get_creation. Call it with that ID instead of starting a new generation: a new request is a new charge.