MCP server
Limits & what it can't do
The honest list — what the Analyze AI MCP server doesn't do, plus the rate limits and caps that keep it safe and affordable.
It reads your tracked data — it doesn't invent data
The MCP server answers from what your workspace has already measured: your tracked prompts, their runs across the AI engines, and the mentions, sentiment, and citations extracted from them.
- If you don't track it, it can't see it. Ask about a topic outside your prompt portfolio and the correct answer is "no data" — the fix is to add prompts for it, which you can do right from the chat with write access.
- Freshness follows your run schedule. Answers reflect the latest completed runs, not a live crawl of the engines at question time. You can trigger an on-demand run when you need fresh numbers.
- It never scrapes the web. Sources, authority scores, and citations come from your workspace's citation intelligence, not ad-hoc browsing.
What's out of scope today
- Writes are limited to prompt management. Creating, editing, tracking, and untracking prompts plus triggering runs — that's the full write surface. Brand Vault, competitors, content tools, team, and billing are dashboard-only by design.
- Traffic analytics aren't exposed. Page-level session and engagement data stays in the dashboard for now; the MCP focuses on visibility, perception, and citation intelligence.
- Strategy narratives come in fixed windows. The pre-computed narrative themes are available for 7, 14, 30, and 90-day windows. Custom ranges still get full metrics — just not the narrative layer.
- One workspace per connection. There's no cross-workspace rollup; agencies connect each client workspace separately.
- No streaming. Long analyses return when complete rather than progressively — expect a short wait on heavyweight questions.
Rate limits & caps
These keep a runaway agent from burning your quota (each on-demand run queries six AI engines in real time):
| Limit | Value |
|---|---|
| Requests per token | 60 / minute |
| Requests per workspace | 120 / minute |
| On-demand prompt runs | ~2 per 10 minutes |
| Prompt changes per request | 10 |
| Evidence references resolved per request | 20 |
Hitting a limit isn't an error state — the assistant is told exactly how long to wait and typically retries on its own.
Plan boundaries
- The MCP server requires the Growth plan; connections on lapsed plans stop working at the next request.
- Prompts created over MCP count against the same tracked-prompt quota as the dashboard — the server checks your allowance before writing, so you can't overshoot by accident.
Something you need that's on this list? Tell us — the write surface and exposed analyses grow based on what customers ask their assistants to do. Reach us from the dashboard or at support.