Your agent builds the site. We tell it exactly what to fix.

Quickstart

1. 1

   Create an account on a paid plan

   API access is included on paid plans — it starts on Monitor. Sign up, then add your website and run your first scan so there is a plan to pull.

2. 2

   Create an API key in Settings

   In the dashboard, go to Settings → API keys. Keys start ams_, carry the scopes you pick (read the plan, mark items done, verify fixes), and are shown once — we store only a hash.

3. 3

   Hand it to your agent

   Set the key as the AMS_API_KEY environment variable and give your agent the skill bundle — or just paste the two calls below.

4. 4

   Run the loop

   Pull the plan, implement the top-ranked item in your codebase, deploy, verify, mark it done, repeat. Ranks re-flow as items close.

Two calls to see your plan

Everything keys off the domain id from the first call — agents should never guess ids or pass URLs.

export AMS_API_KEY=ams_...   # from Settings → API keys

# 1. List your domains (always first — everything keys off the domain id)
curl -s https://app.goaimysite.com/api/v1/domains \
  -H "Authorization: Bearer $AMS_API_KEY"

# 2. Pull the ranked action plan for a domain
curl -s "https://app.goaimysite.com/api/v1/action-plan?domain_id=dom_12" \
  -H "Authorization: Bearer $AMS_API_KEY"

The plan comes back as one ranked queue: each item has a title, the page it applies to, the flagged value the scan found, a copy-ready recommended value where possible, and machine-checkable acceptance criteria.

The verify loop — proof, not promises

After a fix deploys, one call fetches your live page (only your own registered site is ever fetched) and compares what is actually there against what the scan flagged:

# After your agent deploys a fix, prove it landed:
curl -s -X POST https://app.goaimysite.com/api/v1/verify \
  -H "Authorization: Bearer $AMS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"item_ids": ["rec_123"]}'

# verdict "changed"  → the live page no longer shows the flagged value: mark it done
curl -s -X POST https://app.goaimysite.com/api/v1/action-plan/items/rec_123/complete \
  -H "Authorization: Bearer $AMS_API_KEY"

• changed — the live page no longer matches the flagged value: the fix is live, mark the item complete.
• unchanged — the flagged value is still there: the deploy has not landed yet (or a cache is stale). Wait and re-verify.
• not_checkable — no single on-page signal exists for that item type; the response says why.

Verification is a fast, deterministic page check — not a re-scan. Scores and AI visibility update on your next scheduled scan.

Prefer MCP? Add it as a connector

The same API is available as a remote MCP server, so agents that speak the Model Context Protocol get the five calls as native tools — list_domains, get_action_plan, get_item, complete_item and verify_items — with the same key, scopes and rate limits as the REST endpoints.

https://mcp.goaimysite.com

Add that URL as a connector (custom MCP server) in your agent — that is the whole setup. Your client discovers the sign-in flow on its own and opens a browser window where you approve access to your account; there is no key to create, copy or paste, and you can withdraw access at any time.

For headless use — CI, scripts, an agent with no browser — send an ams_ API key as the bearer token instead. Both routes hit the same tools with the same scopes, rate limits and paid-plan requirement. Stateless Streamable HTTP either way.

The older address https://app.goaimysite.com/api/v1/mcp still works with an API key, so existing connectors keep running. Use mcp.goaimysite.com for browser sign-in.

Machine-readable surfaces

• MCP server — the same five calls as native MCP tools, over Streamable HTTP
• OpenAPI 3.1 spec — every endpoint, schema and error code, written for agents
• Claude Skill bundle — SKILL.md teaching the full fix-and-verify loop, with exact curl calls
• Plugin manifest — bearer-auth manifest pointing at the OpenAPI spec
• llms.txt — plain-text map of these surfaces for agents

Rate limits per key: 60 reads/min, 10 writes/min. Errors use one envelope — { "error": { "code", "message", "details"? } } — and 401/403 are permanent, so agents are told not to retry them.