API

Test your site with real agents

Send real agents through your site to see what works and where they get lost, then audit it to learn why.

01 Β· connect

Add ora where you work

Your coding agent calls ora over MCP. Your terminal and CI call it through ax. Same journeys, same scans.

In your coding agent
terminal
claude mcp add --transport http ora https://ora.ai/api/mcp
No MCP config? Paste the skill to your agent: https://ora.ai/skill.md
In your terminal
terminal
# send a real agent at the site on one of ora's tasks, free
npx ax journey --domain your-site.com
# score the site and list what to fix
npx ax audit your-site.com
npx fetches it, or npm install -g ax. A goal of your own or a set of journeys takes an ora sign-in, ax login; an ora partner key also runs a goal of your own.

02 Β· journey

Send a real agent through your site

A real agent gets a task, like finding your API docs. You see every page it opened, what it reasoned along the way, and whether it finished.

  • Claude Code
  • ChatGPT
  • eve
  • OpenClaw
  • Gemini
  • Grok
  • Hermes
Claude Code, ChatGPT and eve run on any site; OpenClaw, Gemini, Grok and Hermes once you claim yours on ora.
example output
$ npx ax journey --domain your-site.com --intent api-docs
β–Ά your-site.com Β· claude-code
Β 
β—† Find the API docs and how to authenticate.
β”‚ (…) I'll search for the API docs first.
β”œβ”€β”€ πŸ” "your-site.com API authentication"
β”‚ (…) The docs link from the homepage looks right.
╰── 🌐 your-site.com/ βœ“ 200
╰── 🌐 your-site.com/docs βœ“ 200
β”œβ”€β”€ 🌐 your-site.com/docs/api βœ“ 200
β”‚ (…) This page covers API keys and bearer tokens.
╰── 🌐 your-site.com/docs/api/authentication βœ“ 200
Β 
βœ“ task satisfied Β· your-site.com
8 steps Β· 1 search Β· 24.6s
One of ora's curated tasks, free, no key. A goal of your own, in quotes, runs on ora's platform with an ora sign-in, ax login, or on the public journeys with an ora partner key passed as --api-key.

03 Β· count

Run it many times and find where agents get lost

One run can be luck. A set sends your goals to several agents at several levels of help, repeated, and ora grades and counts every run. The level where agents start to finish shows where your site loses them.

CLI
ax journey "<goal>" --levels
a real set
$ npx ax journey "Find how to create an issue from the API" \ "Find how to receive a webhook from Linear when an issue is updated." \ --domain linear.app --agents claude-code,codex \ --levels --exact https://linear.app/developers
Launch 10 runs? [y/N] y
linear.app set 593eea5e-c76d-4394-8dd7-dc52c6fbb61d
Β 
Goal 1: Find how to create an issue from the API
Level claude-code/claude-haiku-4-5-20251001 codex/gpt-5.4-mini
Need only 0 of 1 0 of 1
claude-code/claude-haiku-4-5-20251001: 1 never visited your site
codex/gpt-5.4-mini: 1 never visited your site
Brand named 0 of 1 1 of 1
codex/gpt-5.4-mini: 1 never visited your site
Exact target 0 of 1 1 of 1
Β 
Goal 2: Find how to receive a webhook from Linear when an issue is updated.
Level claude-code/claude-haiku-4-5-20251001 codex/gpt-5.4-mini
Brand named 1 of 1 1 of 1
Exact target 1 of 1 1 of 1
Β 
linear.app loses agents at the Discovery layer: 0 of 2 runs satisfied at Need only, 3 of 4 at Brand named. An early signal: not enough runs to decide.
Goal 2: The goal names Linear, so the need level is skipped. Write the goal without it to include that level.
No --docs URL, so the docs level is skipped. Add --docs <url> to include it.
At the need level the prompt does not name your site, so an agent can reach the goal without using it.
Next: ax audit linear.app
ora runs, grades and counts the runs; ax previews the set, asks, and prints this. A set takes an ora sign-in on ora Study and ax 0.12.

04 Β· scan

Score the site and see what fails

A journey shows where agents get lost. A scan shows why: it reads the site the way an agent does, scores it out of 100, and lists every failing check with its fix.

MCP
scan_domain
CLI
ax audit
example output
$ npx ax audit your-site.com
your-site.com 58/100 C
Discovery 7/12 passed
Access 22/40 passed
Usability 16/39 passed
Payments 0/6 passed
Top fixes (ranked by ora)
1. openapi-spec β‰ˆ +4 pts
2. llms-txt β‰ˆ +3.2 pts
3. ard-catalog β‰ˆ +1.2 pts
Next: ax journey --domain your-site.com # send a real agent at the site on a task, then read its result

05 Β· fix

Fix what fails, then run it again

Your agent works through the fixes and edits your code, with the skill walking it through each check. A rescan shows what moved, and the next journey shows whether agents now finish.

MCP
get_skill
generate_agent_files
CLI
ax skill
ax generate-files
example output66
$ npx ax audit your-site.com
βœ“ openapi-spec public/openapi.json
βœ“ llms-txt public/llms.txt
βœ“ ard-catalog public/.well-known/ard.json
your-site.com 66/100 C
Working on localhost? ax runs the tunnel and audits the URL it prints:
terminal
npx ax audit localhost:3000 \
  --tunnel-cmd 'ngrok http 3000 --log stdout'
Over MCP, start any tunnel (cloudflared, ngrok, Tailscale) and give your agent its URL.

Reference

Every step, both ways

One journey
MCPrun_journeyget_journey_run
CLIax journey --domain <site>ax journey "<goal>" --domain <site>
A set of journeys
MCPCLI only
CLIax journey "<goal>" --levels
Score a site
MCPscan_domainget_score
CLIax audit <url>
Gate CI on a score
MCPCLI only
CLIax audit <url> --min-score 70
Fix guidance
MCPget_skill
CLIax skill
Draft agent files
MCPgenerate_agent_files
CLIax generate-files <domain>
WebMCP in a real browser
MCPCLI only
CLIax webmcp-audit <url>

FAQ

Before you connect

Is it free?
Free up to a daily limit. With no key or signup, each IP gets 30 scans and 20 of ora's curated journeys a day, at most 10 scans a minute, and reading a score ora already has never counts. Sign up with ax login and your own goals and sets of journeys run on ora Study, whose free plan gives a one-time allowance of 200 runs, up to 100 in one set. Need a higher limit? Contact us.
What can ora see?
Public URLs only, fetched from outside the way an agent would. Code and anything behind a login are out of reach. A tunnel is public only while its process runs.
Why did my score change when I deployed nothing?
The checks moved. New protocols get added and weights shift as agent behavior changes. Rescanning scores you against the current set.
MCP or CLI?
Same work. MCP keeps the tools in every agent session. The CLI fits terminals, scripts and CI, and it alone runs sets of journeys and the WebMCP browser audit.
βœ“ journey complete

Your turn

View API docs β†’