# Market data for AI agents

> Give an autonomous AI agent a crypto market-data toolset — MCP tools with JSON schemas, reports written for LLMs, x402 pay-per-call and machine-readable discovery.

Source: https://www.f1ow.com/docs/use-cases/agent-builders

f1ow is built to be called by models, not just people. Every tool is designed so an agent can pick it, call it correctly and reason over the answer.

## Why it works well for agents

- **One endpoint, many sources.** One MCP server stands in for a dozen market-data APIs. Your agent doesn't juggle keys, rate limits or formats.
- **Reports written for a model.** Over MCP, each result's `content` is a compact report with units, timeframes, coverage and caveats stated, not raw JSON for the model to decode. The same result as JSON (`structuredContent`) is there when your code needs it, and it is what REST and WebSocket return.
- **Precise schemas.** Every tool publishes a JSON input schema with enums, bounds and defaults, and an output schema. Agents call them correctly the first time.
- **Composites.** [Token Dossier](https://www.f1ow.com/docs/tools/research_token_view), [Positioning](https://www.f1ow.com/docs/tools/research_positioning) and [Market Regime](https://www.f1ow.com/docs/tools/research_regime) answer in one call what would otherwise take ten, which saves tokens and turns.
- **Honest outputs.** Responses state what is missing, how fresh each input is and what is modeled rather than observed, so the agent can qualify its answer.
- **Read-only.** Every tool is annotated `readOnlyHint`. Nothing can move funds or place orders, so it is safe to auto-approve.

## Pay per call with x402

An autonomous agent with a wallet needs no account. It calls a priced tool at https://mcp.f1ow.com/x402, receives an MCP `isError` tool result with the payment challenge, and retries with a signed USDC payment on Base in `_meta["x402/payment"]`. REST uses HTTP `402` and payment headers instead. See [x402](https://www.f1ow.com/docs/auth#x402).

## Discovery

Agents and registries can find and describe f1ow on their own:

- [`/.well-known/mcp.json`](https://mcp.f1ow.com/.well-known/mcp.json): endpoints, auth and pricing models
- [`/.well-known/api-catalog`](https://mcp.f1ow.com/.well-known/api-catalog): RFC 9727 linkset to the OpenAPI and AsyncAPI specs
- [`/llms.txt`](https://www.f1ow.com/llms.txt) and [`/llms-full.txt`](https://www.f1ow.com/llms-full.txt): the docs for LLMs
- Each docs page as Markdown: append `.md` to its URL

## Frameworks

Any framework with an MCP client connects in a few lines: the OpenAI Agents SDK, LangChain / LangGraph MCP adapters, the Vercel AI SDK, Microsoft's Agent Framework, or the official MCP SDKs. See [custom agents](https://www.f1ow.com/docs/connect/custom-agent) for the protocol and a TypeScript example. Prefer plain HTTP? Wrap the [REST API](https://www.f1ow.com/docs/api/rest) as function tools using the OpenAPI spec.

## A starter system prompt

```text
You are a crypto market analyst. Use the f1ow tools for any market fact.
Start broad (research_regime), then narrow to a coin (research_positioning,
ta_technicals, flow_levels). Check generatedAt and coverage before stating a
number, and say when an input is modeled or missing. Never present a setup as
a recommendation; describe the evidence and the risk.
```

> Get a key for your agent: https://www.f1ow.com/dashboard