Use reAPI from AI Agents

An agent can drive reAPI three ways. Pick one; they bill the same and share the same API key, limits and models.

Want the agent itself to think on reAPI's chat models instead? See Coding agents on reAPI.

Set your key once

Create a key under Dashboard → API Keys and export it, so the agent can read it without the key landing in your project files:

export REAPI_API_KEY=YOUR_API_KEY

The API Keys page also has a Connect via MCP button that prints a ready config for each MCP client.

What an agent does on reAPI

1. Find a model — list_models (MCP), or the model list in /llms.txt. Every model has a plain-text summary at https://reapi.ai/models/<slug>/llms.txt with its parameters, prices and examples.
2. Read its input schema — get_model_schema (MCP), or the model's docs page.
3. Use public media URLs — generation endpoints accept only public http(s) URLs for images, video and audio; base64 and data: URIs are rejected, so host local files somewhere reachable first.
4. Check the price — estimate_cost (MCP) returns the credits a request would cost before anything is submitted. 1 credit = $0.001.
5. Submit and wait — submit_task, then get_task with wait_seconds: 50 until completed or failed. Over HTTP, POST the model's endpoint and poll GET /api/v1/tasks/{id} every 1–2 seconds.

Failed tasks are refunded automatically, except where a model's docs say a blocked output is still charged; the settled charge is usage.credits on the task. A submit is not idempotent: retrying a request whose response you did not receive creates and bills a second task, so check the task list or balance first.

Limits to plan around

On HTTP 429, wait for Retry-After when it is present, otherwise back off exponentially with jitter and a cap. Key off the status rather than parsing error bodies, which follow each protocol (chat endpoints use OpenAI or Anthropic error shapes). When task submits hit 50002, wait for running work to finish before submitting more — a batch job should keep a queue rather than fire everything at once. Contact support if you need higher limits.

Machine-readable entry points