# Quick Start: Agent Endpoints > Create and call your first AI agent endpoint in 3 minutes. Create your first AI agent endpoint and call it from the command line. This takes about 3 minutes. You need an Aerostack account and at least one MCP workspace with a connected server. If you have not set up a workspace yet, see [MCP Workspaces](/workspaces). 1. **Create the Agent Endpoint** **From the Dashboard:** 1. Log in to the [Aerostack Dashboard](https://app.aerostack.dev) 2. Navigate to **Agent Endpoints** in the sidebar 3. Click **Create Endpoint** 4. Fill in the form: - **Name:** `summarizer` - **System Prompt:** `You are a concise summarizer. Given any text, produce a clear 2-3 sentence summary that captures the key points.` - **Workspace:** Select your workspace - **LLM Model:** `gpt-4o-mini` (default) - **Output Format:** `text` 5. Click **Create** You will see your API key and run URL. Copy the API key now — it is only shown once. **Or via the API:** ```bash curl -X POST https://api.aerostack.dev/api/agent-endpoints \ -H "Authorization: Bearer YOUR_JWT_TOKEN" \ -H "Content-Type: application/json" \ -d '{ "name": "summarizer", "workspace_id": "ws_your_workspace_id", "system_prompt": "You are a concise summarizer. Given any text, produce a clear 2-3 sentence summary that captures the key points.", "llm_model": "gpt-4o-mini", "output_format": "text" }' ``` Response: ```json { "endpoint": { "id": "aep_7f3a2b9c1d4e5f6a8b", "name": "summarizer", "slug": "summarizer", "status": "active", "auth_mode": "api_key", "llm_provider": "openai", "llm_model": "gpt-4o-mini", "output_format": "text", "max_tool_calls": 10, "timeout_ms": 30000, "temperature": 0.3, "max_tokens": 4096, "price_per_run_cents": 0 }, "api_key": "aek_a1b2c3d4e5f6a7b8c9d0e1f2a3b4c5d6", "run_url": "/api/run/summarizer" } ``` Save the `api_key` value immediately. It is only returned once during creation. If you lose it, use the regenerate-key endpoint to get a new one (which invalidates the old key). 2. **Call Your Agent** Use the `run_url` and `api_key` from the previous step: ```bash curl -X POST https://api.aerostack.dev/api/run/summarizer \ -H "Authorization: Bearer aek_a1b2c3d4e5f6a7b8c9d0e1f2a3b4c5d6" \ -H "Content-Type: application/json" \ -d '{ "input": "Aerostack is a B2B2X API platform that serves as a headless backend builder for developers. It allows teams to create, configure, and monetize backend APIs with built-in CMS, authentication, e-commerce, AI capabilities, real-time features, and an API gateway. The entire platform runs on Cloudflare edge infrastructure for low-latency global performance. Developers can deploy functions, connect MCP servers for tool access, and publish APIs to a community marketplace." }' ``` Response: ```json { "output": "Aerostack is a developer-focused API platform that combines CMS, auth, AI, real-time, and API gateway features into a single headless backend builder. Built on Cloudflare's edge network, it enables teams to create, monetize, and deploy backend APIs with low-latency global performance, including MCP tool integration and a community marketplace.", "usage": { "tokens_input": 312, "tokens_output": 67, "cost_cents": 0.08, "latency_ms": 1456, "iterations": 1 } } ``` 3. **View Execution History** Check the runs for your endpoint in the dashboard, or via the API: ```bash curl https://api.aerostack.dev/api/agent-endpoints/aep_7f3a2b9c1d4e5f6a8b/runs \ -H "Authorization: Bearer YOUR_JWT_TOKEN" ``` ```json { "runs": [ { "id": "run_8e4f2a1b3c5d7e9f0a", "endpoint_id": "aep_7f3a2b9c1d4e5f6a8b", "input": "Aerostack is a B2B2X API platform...", "output": "Aerostack is a developer-focused API platform...", "tokens_input": 312, "tokens_output": 67, "cost_cents": 0.08, "latency_ms": 1456, "status": "success", "created_at": "2026-03-15T10:23:45Z" } ] } ``` ## Test from the Dashboard You can also test your endpoint directly from the dashboard without using the API key. The dashboard uses your JWT session for authentication: ```bash curl -X POST https://api.aerostack.dev/api/agent-endpoints/aep_7f3a2b9c1d4e5f6a8b/test \ -H "Authorization: Bearer YOUR_JWT_TOKEN" \ -H "Content-Type: application/json" \ -d '{"input": "Test input text here"}' ``` Test runs are rate limited to 10 per minute per endpoint and logged with `status: "test"` in the run history. ## Input Field Flexibility The execution endpoint accepts input in several field names for convenience: ```json {"input": "your text"} {"message": "your text"} {"text": "your text"} {"prompt": "your text"} ``` All are equivalent. If none of these fields are present, the entire request body is stringified and used as input. ## What Happens Behind the Scenes When you call `/api/run/:slug`, the platform: 1. **Authenticates** — Checks the API key or allows public access 2. **Rate limits** — 60 requests per minute per endpoint; 20 per minute per IP for public endpoints 3. **Loads MCP tools** — Loads all available tool definitions from the workspace 4. **Runs the agent** — Sends your input + system prompt to the LLM with available tools 5. **Executes tool calls** — If the LLM requests tools, executes them and feeds results back 6. **Repeats** — The loop continues until the LLM produces a final response or hits the `max_tool_calls` limit 7. **Charges** — If `price_per_run_cents > 0`, deducts from the owner's wallet 8. **Logs** — Records the run with input, output, token counts, cost, and latency ## Next Steps - [SSE Streaming](/agent-endpoints/streaming) — Get real-time progress events during execution - [Output Formats](/agent-endpoints/output-formats) — Return structured JSON instead of text - [Pricing](/agent-endpoints/pricing) — Set per-run charges and understand cost breakdown - [API Reference](/agent-endpoints/api-reference) — Full endpoint and schema documentation