Prerequisites
Before you start, make sure you have:- An AIsa API key (console.aisa.one/api-keys)
- An agent codebase using any language or framework compatible with the OpenAI or Anthropic API
1. Point your agent’s LLM calls at AIsa
Store your API key in an environment variable:model string to switch providers without needing additional SDKs or credentials. Agent frameworks (such as LangChain, the Vercel AI SDK, and LlamaIndex) work the same way: set their OpenAI-compatible base URL to https://api.aisa.one/v1.
2. Give your agent AIsa tools over MCP
If your framework includes an MCP client, connect it to the AIsa MCP server (Streamable HTTP):- OAuth (recommended when the agent acts on behalf of a signed-in user): The server publishes standard OAuth metadata at
https://mcp.aisa.one/.well-known/oauth-protected-resource/mcp, allowing an MCP client with OAuth support to manage browser sign-in and token storage. - API key (for a server-side agent you operate): Pass
Authorization: Bearer $AISA_API_KEYon every MCP request using the same key from step 1.
search and list_categories to find operations, get_details (also served as batch_get_schema) to inspect operation schemas, batch_quote to price the exact calls without running them, and batch_use to run up to 20 calls. All but batch_use are free; batch_use is billed per call. use remains as a deprecated single-call form. No per-endpoint tool definitions are needed.
If your framework lacks an MCP client, call the Data APIs directly over HTTP using the same key. In either setup, teach your agent AIsa’s rules for quotes and approval by including https://aisa.one/docs/agent-quickstart.md in its system prompt.
Try it end to end:
Quick reference
Troubleshooting
Related
Agent Quickstart
The agent-facing instruction file this tutorial builds on.
Authentication
API key creation, rotation, and storage best practices.
Model Catalog
Every model ID, context window, and endpoint your agent can call.