Contentstack launches Canoe to track brands in AI answers
Canoe tracks brands in AI answers and connects reports to agents with MCP permission checks
Contentstack announced Canoe on October 1 to monitor how AI answers describe brands. It works with sites outside Contentstack’s publishing platform. Partner XCentium used Canoe in early access, but the announcement does not date that earlier availability.
Review the questions first
The setup guide asks users to review a drafted business profile. Growth subscribers can change the suggested questions before creating a monitor.
Reports include mentions, descriptions and citations, but the numbers come from different methods. According to the metrics documentation, visibility and citation counts are calculated in code. Sentiment, market position and the brand leaderboard involve model judgments. The overall score blends counted and judged inputs.
A passing reference counts toward visibility just as a recommendation does. Free and Growth reports also cover different engines, so their overall scores should not be compared directly. A partially completed run can change that comparison again because it includes only the engines that answered.
The free plan has a small reporting budget
The plan limits allow 20 prompts a month on Free across ChatGPT and Gemini. One question sent to one engine consumes one prompt. A report containing ten questions across those two engines therefore uses the whole monthly allowance. The option to request another report does not provide extra capacity.
Growth adds Perplexity and Claude and 2,000 monthly prompts. Its advertised launch price is $279 a month or $2,790 a year. Enterprise is planned for early 2027.
Agent access needs its own permission check
The paid MCP server lets agents read reports and request new runs. Website edits need separate CMS or repository tools and the team’s permission.
There is a setup detail worth checking before connecting an agent. MCP currently requires an API key with both read and write access, even when the intended job is only reading reports. The documentation recommends blocking the run trigger when new runs should not be allowed. Reading a report does not spend prompts, while starting a run does.
OAuth sign in and read only MCP keys are still listed as coming soon. The docs consequently list the ChatGPT connector as unavailable for now, while clients that send the required key header have a supported connection path.
Read the answer before acting on the score
For a team trying Canoe, the first useful test is whether its selected questions and cited answers reveal a real information gap. Keep the question set and engine coverage consistent when comparing runs. A higher visibility score alone cannot establish that a content change improved sales.
Archival photograph Statistics on a laptop by Carlos Muza, published on April 17 2016, under the Unsplash License. The image illustrates reporting software and does not show Canoe.
