Agent access (MCP)

1. Make a token

In the dashboard, open Agent access and create a token. It starts with kr_pat_ and is shown once: we keep only a hash, so we can’t show it again. Revoke it there at any time and it stops working at once. You can have 10 active tokens.

2. Connect

The server speaks MCP over Streamable HTTP, JSON replies, at:

https://keywordrace.com/mcp

Send the token as a bearer token. With Claude Code:

claude mcp add --transport http keyword-race https://keywordrace.com/mcp \
  --header "Authorization: Bearer kr_pat_YOUR_TOKEN"

Or in any client that takes a JSON config:

{
  "mcpServers": {
    "keyword-race": {
      "type": "http",
      "url": "https://keywordrace.com/mcp",
      "headers": { "Authorization": "Bearer kr_pat_YOUR_TOKEN" }
    }
  }
}

To check it by hand:

curl https://keywordrace.com/mcp \
  -H "Authorization: Bearer kr_pat_YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Tools

• check_keyword. A keyword in one storefront: top competing apps, estimated demand and competition, related keywords, and your app's position if you pass it.
• related_keywords. Related keyword suggestions for a keyword or an app.
• audit_listing. The 0 to 100 listing audit (an estimate) with the top fixes and a shareable path.
• list_keyword_lists. Your keyword lists.
• get_keyword_list. One list with each keyword's latest position, 7-day change and estimates.
• create_keyword_list. A new list for a storefront, optionally for one of your apps.
• add_keywords. Add keywords to a list. Stops at your plan's cap.
• remove_keyword. Remove one keyword from a list.
• delete_keyword_list. Delete a list and its keywords.
• list_listings. Your listings: source text of a store listing, kept to be translated.
• get_listing. One listing: source text, character limits, supported languages and the drafts so far. Read it before writing.
• create_listing. Start a listing from an app's public name and description, or from text you give.
• save_listing_draft. Save text you wrote for one language as a draft. Text over the App Store's limits is refused with the reasons.

Storefronts are two-letter codes such as us, gb or de; 50 are supported. Apps are App Store links or numeric ids.

Writing translations with your agent

Your agent can write your store listing in 15 languages without any key on our side: it calls get_listing, writes each language, and saves it with save_listing_draft. Every saved text is only a draft. You read, edit and approve it on the site; the agent has no tool to approve or send anything.

Limits

• Each token: 30 calls a minute and 1,000 a day. Over that you get HTTP 429 with a Retry-After header.
• Lookups to Apple are shared with everyone and limited to a fixed number a minute. When it’s used up, a tool returns an error saying so; wait a minute and ask again. Results are cached for a day, so repeats are free.
• An audit reads up to 6 of Apple’s public lookups the first time, and none if the app was audited in the last day.
• A token acts as you, within your plan: the same list and keyword caps as on the site.

What the numbers mean

Positions are positions in Apple’s search API results, which may differ from what a phone shows. Demand, competition, related keywords and the audit score are estimates we work out from public app data. None of them is search volume, and the audit can’t see the keyword field in App Store Connect. App Store only: there is no Google Play data.

Errors

A missing, wrong or revoked token gets HTTP 401. A tool that fails (Apple busy, an unknown list, a bad keyword) returns a normal MCP result with isError true and a sentence your agent can read. Treat the token like a password: anyone who has it can use your lists.