// free agent readiness check
Can an AI agentsign up and use your service?
An AI Assistant has to discover a service, read its API and tools, register, authorise and connect, all without a person. Enter a domain. You get a score out of 100 and a fix for every item that is not a pass.
Read-only. Public URLs only, plus MCP metadata handshakes. Takes about 10 seconds.
// what we check
The four steps an agent takes.
Discover 20 points
The home page answers over HTTPS. /llms.txt says what the service does. robots.txt does not block GPTBot, ClaudeBot, PerplexityBot and the other agents this check names.
Understand 25 points
An OpenAPI file sits at a standard path and names how to authenticate. If an MCP server lists tools, each one has a description, an input schema and a read-only or destructive hint.
Sign up and authorise 30 points
OAuth metadata is published. Dynamic client registration is advertised, so an agent can get a client ID without a person. Protected-resource metadata names the authorization server.
Connect 25 points
A remote MCP server answers initialize and tools/list, or a 401 tells the agent where to authorise. The check tries /mcp, /api/mcp, the api. and mcp. subdomains, and a URL listed in llms.txt. No tool is called.
// the score
Same site, same score.
A pass counts in full, a warning counts as half, and a fail counts as zero. The rules do not change between runs.
// the report
What you leave with.
A score and four stage totals
One number out of 100, then Discover, Understand, Sign up and authorise, and Connect against their own weights.
Pass, warning or fail, with the URL
Every item says what was found. When a file or endpoint answered, the report names that URL.
A fix list, then a diff
Anything that is not a pass has a fix. The next check of the same site shows which items changed.
Most fixes are a file at a well-known path. If there is no API or MCP server to publish, Build makes one. A Reliability plan re-runs this journey and emails you when a pass becomes a fail.
// questions
About the check.
What does the Agent Readiness Check test?
Whether an AI agent can discover your service, understand its API and MCP tools, register and authorise itself, and connect without a human in the loop. It reads public well-known URLs such as llms.txt, robots.txt, OpenAPI files and OAuth metadata, then sends MCP initialize and tools/list handshakes.
Does it change anything on my site or call my API?
No. It only sends read-only requests to public URLs and MCP metadata handshakes. It never registers a client, never sends credentials and never calls a tool.
How is the score calculated?
Four stages with fixed weights: Discover 20, Understand 25, Sign up and authorise 30, Connect 25. Each item counts as pass, half for a warning, or zero. The rules are deterministic, so the same site gives the same score.
What happens after the check?
You get a fix for every item that is not a pass. If there is no API or MCP server yet, a Build creates one. A Reliability plan then re-runs the same journey on a schedule and alerts you when a pass becomes a fail.