Overwing Beacon
Is your product reachable by agents?
More and more of the visitors choosing software are agents. They do not see your landing page the way a person does: they look for an llms.txt, an MCP server, an agent card, an API description, and a page they can read without a browser. Beacon makes the same requests and tells you what it found, what is missing, and what to change first.
See what you get first: a real report on our own site.
Three questions
Find, read, use. In the order an agent meets you.
Find
Can an agent discover the product and where its surfaces are?
- robots.txt lets AI agents in
- A sitemap lists the pages
- llms.txt tells a model what the product is and where to go
- The MCP server is in the official MCP Registry
Read
Can an agent read what the product is?
- The home page is served to an automated client
- The page has content without running JavaScript
- The page has a title and a description
- The page carries structured data
- A markdown version of the page is offered
- The text says what the product does · judged
Use
Can an agent call the product without a person?
- An MCP server card is published
- A hosted MCP endpoint answers
- An A2A agent card is published
- An OpenAPI document is at a well-known address
- An agent can pay without an account
- The text tells an agent how to start without a person · judged
- The text states the price · judged
Most checks are facts: a file is there or it is not. Three are judgments a program cannot make, such as whether your text says what the product does; those are answered by TypeSafe’s Jev model and marked as judged. The check is read-only, about twenty requests, identified as OverwingBeacon. It never calls a tool on your MCP server and never messages your agent.
For agents
Free with a key. One dollar without one.
An agent with an Overwing key starts the check with POST /api/v1/beacon/checks and reads the full report from the status address. It costs nothing, and the key is free too: POST /api/v1/signup returns one in a single call.
An agent with no key makes the same two calls and gets the summary: the score, the three answers and the first fix.
An agent with a wallet and no account calls one endpoint and pays $1 in USDC over x402. The full report is the response. A site that cannot be checked is not charged.
Over MCP, the tools are beacon_start, beacon_report and beacon_sample on the hosted server at /mcp.
Reports can be kept. Download one as Markdown or JSON, print it to PDF, or save it to your dashboard; an agent adds ?format=md or sends its key.
Every failed check carries a fix written to be acted on, and top_fixes ranks them by how much each is worth, so an agent can open the pull request itself.
Described for machines at /api/v1/beacon, /beacon.md and /llms.txt.
GET /api/x402/beacon?url=yourproduct.com
402 Payment Required # the terms: $1 USDC on Base
…sign with any x402 client and resend…
200 OK
{ "status": "complete", "score": 57, "verdict": "partly",
"categories": [
{ "key": "find", "answer": "yes", "points": 23, "max": 23 },
{ "key": "read", "answer": "yes", "points": 37, "max": 41 },
{ "key": "use", "answer": "no", "points": 3, "max": 47 } ],
"top_fixes": [
{ "check": "a2a_agent_card", "gain": 12,
"fix": "Publish /.well-known/agent-card.json (A2A): name, …" },
{ "check": "mcp_endpoint", "gain": 10,
"fix": "Host an MCP endpoint (Streamable HTTP) so an agent …" } ],
"checks": [ { "id": "llms_txt", "status": "pass", "detail": "…" }, … ] }Pricing
Free. No card, nothing to subscribe to.
The check is free. The fixes can be done for you. If you would rather not make the changes yourself, we do them for a fixed price, as pull requests, and run the check again when they are live. Agent-ready, done for you.