AEO Engine MCP

A company can now be summarised wrongly by four assistants at once and never find out, because the wrong answer is delivered to someone else and nobody forwards it back.

Server type: Remote (Streamable HTTP)
Endpoint:    https://aeoengine.ai/mcp
Auth:        none, keyless
Tools:       2 (run_aeo_report, get_aeo_report_status)
Engines:     ChatGPT, Perplexity, Claude (public profile)
Input:       domain plus a real work email, MX validated
Output:      emailed report plus a shareable report URL
Cost:        free on the public path

Why This Matters for Operators

Search visibility had a referee: the results page. Assistant visibility does not. Each engine answers differently, cites differently and changes without notice, and the only way to know what it says about a business is to ask it the same question repeatedly and record the answer.

This server does that measurement as a job. run_aeo_report takes a domain and an email, runs the engine cohort, and returns a report that breaks the result down per engine with provenance, plus competitor gaps, coverage, methodology, the exact questions that were measured, competitor evidence and a named share of voice. The finished report is emailed to the address supplied and a shareable URL comes back.

The markdown_summary field is the part that matters for an assistant workflow: it is written to be read out to a person, so the agent does not have to invent the narrative from a JSON tree.

What It Returns, Verified

Driven end to end without a key. initialize on https://aeoengine.ai/mcp returned HTTP 200 as aeo-engine-aeo-report 2.0.0 on protocol 2025-03-26, advertising the tools capability. tools/list returned exactly two tools.

run_aeo_report requires domain and email. The schema is unusually explicit for a free surface:

• email is the user's real work address, MX validated, with disposable domains rejected, because the report is sent there
• wait_seconds waits inline up to about 50 seconds; the run itself can take up to roughly ten minutes, so a client that cannot wait polls
• customQuestions takes 5 to 20 exact questions and preserves their text and order verbatim, which is how the same prompt set gets compared month over month
• competitor pins one domain on the public path, while competitors accepts up to five only with an agency key
• context accepts a vertical hint, a city and region, and up to eight focus terms
• report_profile selects the engine cohort, public or aio
• fresh forces a new run instead of a cached report
• idempotency_key is 16 to 128 printable characters and is required for a trusted create, so a retried call maps to one job

get_aeo_report_status polls a submitted job once. The status vocabulary is honest about the middle: queued, preparing, needs_review and running mean keep polling, while complete, partial and failed are terminal, and state carries the phase and the reason. On completion it returns the structured report, the report_url and the markdown_summary.

Two schema details are worth more than a feature list. The company field is off limits in the prompt guidance, which reads as a deliberate guard against an agent inventing an entity. And an unusable context value is rejected with a reason, 422 on the public path or 400 with an agency key, rather than being silently dropped.

Limits Worth Recording

• It measures, it does not fix. The output names coverage gaps and prioritised fixes; nobody edits the site for you.
• One competitor on the free path. Five-way comparison requires an agency key.
• The email is the price. There is no anonymous path to a report, and the report arrives in that mailbox. An agent should not fill it with its own address; the instruction is explicit about the user's real one.
• Cached by default. A repeated call may return a cached report unless fresh is set, so a month-over-month comparison has to force the run.
• Partial is a real outcome. A partial report means some engines answered and others did not. Reading it as a complete picture is the mistake the status field exists to prevent.
• No published pricing for the agency path. The public report is free; anything beyond one competitor, custom questions at scale or more frequent runs goes through an agency key whose terms are not on the server.

Wiring It Up

{
  "mcpServers": {
    "aeo-engine": {
      "url": "https://aeoengine.ai/mcp"
    }
  }
}

Ask for a report on your own domain, give a real address, and poll with the returned job id if the client timeout is under a minute. For a repeatable baseline, send the same customQuestions set every month with fresh: true so the questions and the order never drift.

FAQ

Does it need an API key?

No. Both tools answer on a keyless connection. The report is gated by an email address rather than a credential.

Which engines does it measure?

The public profile covers ChatGPT, Perplexity and Claude. The report_profile parameter selects the cohort, and the report breaks results down per engine with provenance.

How long does a report take?

Up to about ten minutes. wait_seconds can hold the call inline for about 50 seconds, after which the job id has to be polled with get_aeo_report_status.

Can I compare more than one competitor?

One on the public path. Up to five are accepted with an agency key, and the public report pins exactly one.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects