Your report, inside the assistant

What It Reads

Six tools, and every one of them is the same computation the dashboard renders over your stored answers. Ask in your own words; the assistant picks the tool.

• get_overview_metricsThe headline numbers: visibility, sentiment, average position and share of AI voice, under any filter
• compare_runsThis window against the last one, so a delta can be stated, and a plain answer when there is no prior run
• get_question_detailOne question: its visibility, which engines answered it, which brands those answers named, and how many stored answers back that up
• list_lost_questionsEvery question you are losing, with sample counts so a failed call never reads as absence
• get_sources_summaryThe domains the answers cite most, labelled yours, competitor or third-party
• get_remediation_queueThe Fixes queue with status and priority, on the plans that include Fixes

The headline tool reports grounding per engine, and the server tells the assistant at connect time to check it before presenting a number as what that assistant answers today. A failed engine call is never counted as your absence: failed samples leave every metric, and the counts travel with the answer so the assistant can say how many samples it is standing on.

Three Steps to Connect

1. 1. Mint a token

   Settings, then API tokens, on any Track, Grow or Command account. The token is shown once. It reads every property created under your sign-in email, it never expires, and revoking it is the only way to end it. Up to 5 live at a time, 60 calls each per clock hour.

2. 2. Add the server

   Claude Code, one line:

   claude mcp add --transport http aeosearch https://www.aeosearch.io/api/mcp \
     --header "Authorization: Bearer aeo_…"

   Cursor, or any client that reads a JSON config with headers:

   {
     "mcpServers": {
       "aeosearch": {
         "url": "https://www.aeosearch.io/api/mcp",
         "headers": { "Authorization": "Bearer aeo_…" }
       }
     }
   }

   Anything else that speaks JSON-RPC can post straight to the endpoint with the same header. The server answers initialize, tools/list and tools/call, one call per request.

3. 3. Ask

   “Which questions did we lose on ChatGPT last run, and which domains did the winning answers cite?” If the account owns more than one property, name it, or pass property in the call: the error lists your property ids.

What It Never Does

Nothing a token can reach changes anything. The Fixes queue moves only from the dashboard, on purpose: an assistant that can quietly rewrite your remediation list is the kind of thing you find out about later, so we built the read side first and stopped there.

One thing it does not do yet: the connector settings inside claude.ai and ChatGPT expect OAuth, and this server speaks a bearer token. A token pasted there will not authenticate. Claude Code, Cursor and every client that sends a header have what they need; the OAuth door is on the list, and this page will say so the day it opens.