Agent access

AI agents are first-class users. Everything a person can do with a scan, an agent can do through the API or the MCP server, with the same account, the same allowances and the same price.

Discover the product

https://accessdiff.com/llms.txt: a plain-text summary written for language models.
https://accessdiff.com/.well-known/product.json: a machine-readable description of the scan, its inputs and outputs, prices and endpoints.
https://accessdiff.com/api/v1/openapi.json: the OpenAPI 3.1 document, generated from the same schemas the server validates with.

Call the API

Authenticate with an API key and send X-Usage-Source: agent so the account's usage meter attributes the scans to agents rather than to plain API calls. The header is advisory: it changes reporting, never the price.

POST /api/v1/scan
Authorization: Bearer accessdiff_live_…
X-Usage-Source: agent
Content-Type: application/json

{"siteUrl": "https://example.com", "maxPages": 25}

The scan is synchronous and takes one to four minutes; the quick scan (POST /api/v1/quick-scan with {"url": "…"}) answers in about 20 seconds.

MCP server

The product ships an MCP server with two tools, scan and quick-scan, whose input schemas match the API. It runs on the agent's side and calls the API with the agent's key:

{
  "mcpServers": {
    "accessdiff": {
      "command": "npm",
      "args": ["run", "mcp", "--prefix", "/path/to/accessdiff"],
      "env": {
        "PRODUCT_API_URL": "https://accessdiff.com",
        "PRODUCT_API_KEY": "accessdiff_live_…"
      }
    }
  }
}

Results are metered exactly like API calls, with source agent.

Prices for agents

Accounts on a plan use the plan's monthly allowances (Starter 40 site scans and 200 quick scans, Growth 200 and 1,000, Agency 2,000 and 5,000). When an allowance is used up, agent and API calls from an account with a card on file are billed at $2.00 per site scan and $0.05 per quick scan. Free accounts get HTTP 402 with a link to the pricing page; the free tier is the instant scan on the home page.

What an agent gets back

A summary, not a verdict: the score, counts by impact, what changed since the previous scan, the top issues with the failing element and how to fix it, and a signed report link. Automated checks find a portion of accessibility issues; a review by a person is needed to find the rest. Nothing in the result is legal advice, and an agent relaying it should say so.

Rules for agents

Respect Retry-After on 429 responses and 409 conflict (a scan of that site is already running; wait for it).
Do not create more API keys than you need; key management is deliberately session-only.
Read error.message; it is written to be acted on.