Skip to content
legibility

Answers

Straight answers for agent builders.

Direct answers to the questions developers and agents ask about MCP servers, paying per call in USDC with x402, and reading typed product data with a calibrated confidence.

RSS feed · llms.txt

  1. How can an AI agent get typed product data from a URL or a barcode?

    Call read_product with exactly one of a product url or a gtin barcode. Legibility returns a typed product object with a price band, per-field confidence, an overall calibrated confidence, the extraction method and the cost of the call. It is available over REST at /api/v1/read_product and as an MCP tool.

    Updated · product data, read_product, GTIN, API

  2. How can an AI agent pay per API call in USDC?

    An agent pays per call in USDC with x402: it calls the API without a key, receives HTTP 402 with payment requirements, signs a USDC transfer authorization for the quoted amount, and resends the request with an X-PAYMENT header. Legibility's MCP server works this way today on Base Sepolia, a testnet, with no mainnet settlement yet.

    Updated · x402, payments, USDC, MCP

  3. How do I connect an AI agent to a remote MCP server over HTTP?

    Send JSON-RPC 2.0 requests by HTTP POST to the server's endpoint: initialize first, then tools/list to read the tool schemas, then tools/call to run one. Legibility's MCP server is at https://legibility.io/api/mcp, speaks protocol version 2025-06-18, and describes itself at /.well-known/mcp.json. Discovery is free.

    Updated · MCP, agents, JSON-RPC, discovery

  4. What does a calibrated confidence score mean when an agent reads product data?

    A calibrated confidence is a probability: a score of 0.7 means the result is about 70 percent likely to be correct. Legibility fits its scores with isotonic regression on a held-out labelled split and treats 0.7 as the trust gate, so an agent can decide to act, re-check or decline on a number that means something.

    Updated · confidence, calibration, trust gate

  5. Why can't an AI agent read a product page?

    Because a plain request did not return typed facts a machine can extract without guessing. Legibility records which of eight reasons applies: a refusal, a JavaScript-only render, no structured data, not a product page, an extraction below the trust gate, a robots.txt disallow, a timeout, or an inconclusive read.

    Updated · machine readability, structured data, crawlers